Package 'stablehlo'

Title: Write stableHLO programs
Description: The package offers a low level interface to create stableHLO programs. These programs can be compiled and run on different hardware backends (CPU, GPU, ...) using the 'pjrt' package.
Authors: Daniel Falbel [aut] (ORCID: <https://orcid.org/0009-0006-0143-2392>), Sebastian Fischer [aut, cre] (ORCID: <https://orcid.org/0000-0002-9609-3197>), Nikolai German [aut] (ORCID: <https://orcid.org/0009-0001-7394-8367>)
Maintainer: Sebastian Fischer <[email protected]>
License: MIT + file LICENSE
Version: 0.3.0
Built: 2026-06-03 11:50:03 UTC
Source: https://github.com/r-xla/stablehlo

Help Index

Get the last function created

Description

Get the last function created (either via hlo_func or local_func), which is not returned yet.

Usage

.current_func()

Value

A Func object.

Get the current module

Description

Get the current module created via hlo_module or local_module.

Usage

.current_module()

Value

A Module object.

BoolAttr

Description

An attribute holding a boolean value.

Usage

BoolAttr(name, value)

Arguments

name

(character(1))
The name of the attribute.
value

(logical(1))
The boolean value.

Value

BoolAttr

Constant

Description

This represents a constant value.

Usage

Constant(data, type)

Arguments

data

(any)
The value of the constant.
type

(TensorType)
The type of the constant.

Value

Constant

Create a ConstantAttr from R values

Description

Helper function to create a ConstantAttr from R values.

Usage

constant_attr(name, value, dtype = NULL, shape = NULL, simplify_dense = TRUE)

Arguments

name

(character(1))
The name of the attribute.
value

(any)
The R value to convert to a constant.
dtype

(character(1) | NULL)
The dtype of the constant. If NULL, inferred from value.
shape

(integer() | NULL)
The shape of the constant. If NULL, inferred from value.
simplify_dense

(logical(1))
Whether to simplify dense representation. Set to FALSE for multi-dimensional arrays.

Value

(ConstantAttr)

ConstantAttr

Description

An attribute holding a constant value.

Usage

ConstantAttr(name, value, simplify_dense = TRUE)

Arguments

name

(character(1))
The name of the attribute.
value

(Constant)
The value of the attribute.
simplify_dense

(logical(1))
Whether to simplify dense representation. Set to FALSE for multi-dimensional arrays.

Value

(ConstantAttr)

CustomOpBackendConfig

Description

A backend configuration as a list of typed attributes for custom operations. Each element must be a BoolAttr, StringAttr, or ScalarAttr for now. All attribute names must be unique.

Usage

CustomOpBackendConfig(items = list())

Arguments

items

(list)
A list of BoolAttr, StringAttr, or ScalarAttr objects.

Value

CustomOpBackendConfig

DotDimensionNumbers

Description

Represents the dot dimension numbers.

Usage

DotDimensionNumbers(contracting_dims, batching_dims = NULL)

Arguments

contracting_dims

(integer())
The contracting dimensions.
batching_dims

(integer() | NULL)
The batching dimensions.

ErrorConcatenateShapes

Description

Error when input shapes don't match (except at given dimensions)

Usage

error_concatenate_shapes(
  dimensions,
  shapes,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

dimensions

(integer())
The dimensions where shapes may differ (0-based)
shapes

(list() of Shape)
The input shapes
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorDimSizeMismatch

Description

Error when a dimension size doesn't match the expected size at a given index

Usage

error_dim_size_mismatch(
  arg1,
  arg2,
  dim1,
  dim2,
  shape1,
  shape2,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg1

(character(1))
Name of the first argument (e.g. "operand")
arg2

(character(1))
Name of the second argument (e.g. "result")
dim1

(integer(1))
Dimension index in arg1 (0-based)
dim2

(integer(1))
Dimension index in arg2 (0-based)
shape1

(integer())
Complete shape of arg1
shape2

(integer())
Complete shape of arg2
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorDimensionUniqueness

Description

Error when dimension indices are not unique

Usage

error_dimension_uniqueness(
  arg,
  dimensions,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg

(character(1))
Name of the argument that caused the error
dimensions

(integer())
The dimension indices that are not unique.
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorIndexInSet

Description

Error when an index is found in a forbidden set

Usage

error_index_in_set(
  arg1,
  arg2,
  index,
  set,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg1

(character(1))
Name of the argument containing the index
arg2

(character(1))
Name of the argument containing the set
index

(integer(1))
The index that was found in the set
set

(integer())
The set that should not contain the index
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorIndexOutOfBounds

Description

Error when an index is outside the valid range [lower, upper)

Usage

error_index_out_of_bounds(
  arg,
  index,
  lower,
  upper,
  call = sys.call(-1),
  class = character(),
  signal = TRUE
)

Arguments

arg

(character(1))
Name of the argument that caused the error
index

(integer())
The observed index value(s).
lower

(integer(1))
Lower bound of valid range.
upper

(integer(1))
Upper bound of valid range, exclusive.
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorIndicesNotSorted

Description

Error when indices are not sorted in ascending order

Usage

error_indices_not_sorted(
  arg,
  indices,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg

(character(1))
Name of the argument that caused the error
indices

(integer())
The indices that are not sorted.
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorPermuteIndex

Description

Error when permutation values are invalid (not a valid permutation of indices)

Usage

error_permute_index(
  arg,
  permutation,
  expected,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg

(character(1))
Name of the argument that caused the error
permutation

(integer())
The permutation values that are invalid.
expected

(integer())
The expected indices to be permuted.
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorStablehlo

Description

Base error class for all stablehlo errors

Usage

error_stablehlo(
  call = sys.call(-1)[1L],
  ...,
  class = character(),
  signal = TRUE
)

Arguments

call

(call or NULL)
Call that generated the error
...

Additional fields to store in the condition
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorUnequalTypes

Description

Error when types at the same index in two lists don't match

Usage

error_unequal_types(
  arg1,
  arg2,
  index,
  expected,
  actual1,
  actual2,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg1

(character(1))
Name of the first argument
arg2

(character(1))
Name of the second argument
index

(integer(1))
The index where types don't match (0-based)
expected

(character(1))
Description of what was expected
actual1

Type from the first argument (any object with a cli_format method)
actual2

Type from the second argument (any object with a cli_format method)
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

ErrorUnexpectedListType

Description

Error when an element in a list has an unexpected type

Usage

error_unexpected_list_type(
  arg,
  index,
  expected,
  actual,
  call = sys.call(-1)[1L],
  class = character(),
  signal = TRUE
)

Arguments

arg

(character(1))
Name of the argument
index

(integer(1))
The index where the type is unexpected (0-based)
expected

(character(1))
Description of what was expected
actual

(character(1))
What was observed
call

(call or NULL)
Call that generated the error
class

(character())
Additional classes to prepend
signal

(logical(1))
Whether to signal the error (default TRUE)

Format Double Array with Scientific Notation

Description

Formats a double array using scientific notation with 16 digits precision, similar to formatC(x, digits = 16, format = "e").

This is used to embed floating point constants into stableHLO programs.

Usage

format_double(x, precision = 64)

Arguments

x

(double())
Vector to format.
precision

(integer(1))
Currently supports 32 and 64 bit precisions.

Value

character()

Examples

format_double(1.23, 32)
format_double(1.23, 64)

Func

Description

This represents a function. Note: Func uses reference semantics - modifications to a Func object modify the original.

Usage

Func(
  id = FuncId(),
  inputs = FuncInputs(),
  outputs = FuncOutputs(),
  body = FuncBody()
)

Arguments

id

(FuncId
The id of the function.
inputs

(FuncInputs
The inputs of the function.
outputs

(FuncOutputs
The outputs of the function.
body

(FuncBody
The body of the function.

Value

A Func object.

FuncBody

Description

The body of a Func, containing a list of operations.

Usage

FuncBody(items = list())

Arguments

items

(list() of Op)
The operations in the function body.

Value

(FuncBody)

FuncId

Description

This represents the id of a function.

Usage

FuncId(id = "main")

Arguments

id

The id of the function.

FuncInput

Description

This represents an input of a Func.

Usage

FuncInput(id, type, alias = NULL)

Arguments

id

(ValueId)
The id of the input.
type

(ValueType)
The type of the input.
alias

(integer(1) | NULL)
With which output buffer to alias this input.

Value

(FuncInput)

FuncInputs

Description

List of FuncInputs.

Usage

FuncInputs(items = list())

Arguments

items

(list() of FuncInput)
The inputs of the function.

Value

(FuncInputs)

FuncOutput

Description

This represents an output of a Func.

Usage

FuncOutput(type)

Arguments

type

(ValueType)
The type of the output.

Value

(FuncOutput)

FuncOutputs

Description

List of FuncOutputs.

Usage

FuncOutputs(items = list())

Arguments

items

(list() of FuncOutput)
The outputs of the function.

Value

(FuncOutputs)

FuncValue

Description

This represents a variable within a function.

Usage

FuncValue(value_id, value_type, func)

Arguments

value_id

The name of the variable.
value_type

The type of the variable.
func

The function the variable belongs to.

GatherDimensionNumbers

Description

Represents the gather dimension numbers.

Usage

GatherDimensionNumbers(
  offset_dims,
  collapsed_slice_dims,
  operand_batching_dims = integer(),
  start_indices_batching_dims = integer(),
  start_index_map,
  index_vector_dim
)

Arguments

offset_dims

(integer())
The offset dimensions.
collapsed_slice_dims

(integer())
The collapsed slice dimensions.
operand_batching_dims

(integer())
The operand batching dimensions.
start_indices_batching_dims

(integer())
The start indices batching dimensions.
start_index_map

(integer())
Maps start indices to operand dimensions.
index_vector_dim

(integer(1))
The index vector dimension.

Cholesky Operator

Description

See https://openxla.org/stablehlo/spec#cholesky for details.

Usage

hlo_cholesky(operand, lower)

Arguments

operand, lower

(FuncValue)

Details

The values of the other half of the matrix are not guaranteed and backend dependent.

Value

FuncValue

Create a Closure

Description

Creates a new function without any arguments that captures the provided variables.

Usage

hlo_closure(...)

Arguments

...

(FuncValue)
The variables to capture.

Value

(list() of FuncValue)

Examples

func <- local_func()
x <- hlo_input("x", "f32", shape = c(2, 2))
y <- hlo_input("y", "f32", shape = c(2, 2))
f <- hlo_closure(x, y)
print(f)

Create a Constant

Description

Create either a "scalar" (hlo_scalar) or tensor (hlo_tensor) constant. Strictly speaking, stableHLO "scalars" are simply tensors with 0 dimensions. To create an empty constant (at least one dimension is 0), use hlo_empty.

Usage

hlo_scalar(value, ..., dtype = NULL, func = NULL)

hlo_tensor(value, ..., dtype = NULL, shape = NULL, func = NULL)

hlo_empty(dtype, shape, func = NULL)

infer_types_constant(value)

Arguments

value

(any)
Value from which to create a constant.
...

(any)
Additional arguments.
dtype

(character(1))
One of: bool, i8, i16, i32, i64, ui8, ui16, ui32, ui64, f32, f64.
func

(Func)
The function to add the constant to. Per default, uses the last function created with hlo_func or local_func.
shape

(integer())
The shape

Examples

hlo_scalar(1L, dtype = "i32", func = Func())
hlo_scalar(1, dtype = "f32", func = Func())
hlo_scalar(TRUE, func = Func())
hlo_tensor(array(c(1, 2, 3, 4), dim = c(1, 4)), dtype = "f32", func = Func())
hlo_empty(dtype = "f32", shape = c(0, 3), func = Func())

Custom Call Operation

Description

Create a custom call operation that invokes an external function via the FFI (Foreign Function Interface) API.

Note that the attributes called_computations and output_operand_aliases are not implemented yet.

Usage

hlo_custom_call(
  ...,
  call_target_name,
  api_version = 4L,
  has_side_effect,
  backend_config = NULL,
  output_types = NULL,
  operand_layouts = NULL,
  result_layouts = NULL
)

Arguments

...

(FuncValue)
Input values to pass to the custom call.
call_target_name

(character(1))
The name of the registered custom function to call.
api_version

(integer(1))
The API version. Default is 4.
has_side_effect

(logical(1))
Whether the custom call has side effects.
backend_config

(CustomOpBackendConfig | NULL)
Optional backend configuration.
output_types

(list of ValueType | NULL)
The output types of the custom call. Default is NULL (no outputs).
operand_layouts

(list of integer() | NULL)
Layouts for each operand in minor-to-major order. Each element is an integer vector specifying the dimension order. For example, c(0L, 1L) means column-major (dimension 0 varies fastest), while c(1L, 0L) means row-major. Default NULL means no layout constraint.
result_layouts

(list of integer() | NULL)
Layouts for each result in minor-to-major order. Same format as operand_layouts.

Value

(FuncValue | list() | NULL)
The output value(s), or NULL for side-effect only calls.

Create a function

Description

Both functions create a new Func with the given id which is afterwards accessible via .current_func(). Functions receiving a Func as an argument (such as hlo_input, hlo_add, ...) usually use .current_func() by default. You can also directly create a function using Func(), which will not be accessible this way.

Differences between the two functions:

  • local_func removes the function when exiting the current scope, whereas hlo_func does not.

  • hlo_func discards the previously built function(s), whereas local_func does not: after a function created by local_func is either cleaned up automatically (by exiting the scope) or the function is finalized via hlo_return, the previously built function is restored, i.e., accessible via .current_func(). To build nested functions (e.g. to create a closure that is passed to another op), use local_func instead of hlo_func.

Usage

hlo_func(id = "main")

local_func(id = "main", envir = parent.frame())

Arguments

id

(character(1)
The id of the function.
envir

(environment)
Environment where exit handler will be registered for cleaning up the Func if it was not returned yet.

Value

A Func object.

Create a input to a function

Description

Create a input to a function

Usage

hlo_input(name, dtype, shape = integer(), func = .current_func(), alias = NULL)

Arguments

name

(character(1))
The name of the parameter.
dtype

(ValueType)
The data type of the parameter. Can contain digits, letters and underscores. If it starts with a digit, it can only contain digits. Otherwise it must start with a letter.
shape

(integer())
The shape of the parameter. Use integer() for scalars.
func

(Func)
The function id of the parameter. Per default, uses the last function created with hlo_func.
alias

(integer(1) or NULL)
If integer, marks this input as alias with the given output index (0-based).

Examples

func <- hlo_func()
x <- hlo_input("x", "f32", shape = c(2, 2))
print(x)

# You can combine multiple inputs as follows:
c(
  hlo_input("x", "f32", shape = c(2, 2)),
  hlo_input("y", "f32", shape = c(2, 2))
)

Create a module

Description

Both functions create a new Module which is afterwards accessible via .current_module(). Functions created with hlo_func or local_func will automatically register into the current module. The module is finalized when a function named "main" is returned via hlo_return.

Differences between the two functions:

Usage

hlo_module()

local_module(envir = parent.frame())

Arguments

envir

(environment)
Environment where exit handler will be registered for cleaning up the Module if it was not finalized yet.

Value

A Module object.

index_vec

Description

Wraps an integer vector marking it as containing 0-based index values.

Usage

index_vec(x)

Arguments

x

(integer())
Integer vector of indices.

Value

An integer vector with additional class "IndexVector".

Abs Operator

Description

See https://openxla.org/stablehlo/spec#abs for details.

Usage

infer_types_abs(operand)

hlo_abs(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Acos Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloacos_chloacosop for details.

Usage

infer_types_acos(operand)

hlo_acos(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Acosh Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloacosh_chloacoshop for details.

Usage

infer_types_acosh(operand)

hlo_acosh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Add Operator

Description

See https://openxla.org/stablehlo/spec#add for details.

Usage

infer_types_add(lhs, rhs)

hlo_add(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

AfterAll Operator

Description

See https://openxla.org/stablehlo/spec#after_all for details.

Usage

infer_types_after_all(...)

hlo_after_all(...)

Arguments

...

(FuncValue)

Value

FuncValue

And Operator

Description

See https://openxla.org/stablehlo/spec#and for details.

Usage

infer_types_and(lhs, rhs)

hlo_and(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Asin Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloasin_chloasinop for details.

Usage

infer_types_asin(operand)

hlo_asin(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Asinh Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloasinh_chloasinhop for details.

Usage

infer_types_asinh(operand)

hlo_asinh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Atan Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloatan_chloatanop for details.

Usage

infer_types_atan(operand)

hlo_atan(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Atan2 Operator

Description

See https://openxla.org/stablehlo/spec#atan2 for details.

Usage

infer_types_atan2(lhs, rhs)

hlo_atan2(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Atanh Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloatanh_chloatanhop for details.

Usage

infer_types_atanh(operand)

hlo_atanh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

BesselI1e Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlobessel_i1e_chlobessel_i1eop for details.

Usage

infer_types_bessel_i1e(operand)

hlo_bessel_i1e(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

BitcastConvert Operator

Description

See https://openxla.org/stablehlo/spec#bitcast_convert for details.

Usage

infer_types_bitcast_convert(operand, dtype)

hlo_bitcast_convert(operand, dtype)

Arguments

operand, dtype

(FuncValue)

Value

FuncValue

BroadcastInDim Operator

Description

See https://openxla.org/stablehlo/spec#broadcast_in_dim for details.

Usage

infer_types_broadcast_in_dim(operand, broadcast_dimensions, shape)

hlo_broadcast_in_dim(operand, broadcast_dimensions, shape)

Arguments

operand, broadcast_dimensions, shape

(FuncValue)

Value

FuncValue

Call a Function

Description

Calls a named function from within the current function being built. The callee must already be finalized via hlo_return.

Usage

infer_types_call(callee, ...)

hlo_call(callee, ..., simplify = TRUE)

Arguments

callee

(Func)
The function to call. Must be a finalized Func.
...

(FuncValue)
The arguments to pass to the callee.
simplify

(logical(1))
If TRUE (default) and the callee has a single output, return a single FuncValue instead of a list.

Value

FuncValue or list() of FuncValues.

Case Operator

Description

See https://openxla.org/stablehlo/spec#case for details.

Usage

infer_types_case(index, ...)

hlo_case(index, ...)

Arguments

index, ...

(FuncValue)

Value

FuncValue

Cbrt Operator

Description

See https://openxla.org/stablehlo/spec#cbrt for details.

Usage

infer_types_cbrt(operand)

hlo_cbrt(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Ceil Operator

Description

See https://openxla.org/stablehlo/spec#ceil for details.

Usage

infer_types_ceil(operand)

hlo_ceil(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Clamp Operator

Description

See https://openxla.org/stablehlo/spec#clamp for details.

Usage

infer_types_clamp(min, operand, max)

hlo_clamp(min, operand, max)

Arguments

min, operand, max

(FuncValue)

Value

FuncValue

Compare Operator

Description

See https://openxla.org/stablehlo/spec#compare for details.

Usage

infer_types_compare(lhs, rhs, comparison_direction, compare_type)

hlo_compare(lhs, rhs, comparison_direction, compare_type)

Arguments

lhs, rhs, comparison_direction, compare_type

(FuncValue)

Value

FuncValue

Concatenate Operator

Description

See https://openxla.org/stablehlo/spec#concatenate for details.

Usage

infer_types_concatenate(..., dimension)

hlo_concatenate(..., dimension)

Arguments

..., dimension

(FuncValue)

Value

FuncValue

Convert Operator

Description

See https://openxla.org/stablehlo/spec#convert for details.

Usage

infer_types_convert(operand, dtype)

hlo_convert(operand, dtype)

Arguments

operand, dtype

(FuncValue)

Value

FuncValue

Cosh Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlocosh_chlocoshop for details.

Usage

infer_types_cosh(operand)

hlo_cosh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Cosine Operator

Description

See https://openxla.org/stablehlo/spec#cosine for details.

Usage

infer_types_cosine(operand)

hlo_cosine(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

CountLeadingZeros Operator

Description

See https://openxla.org/stablehlo/spec#count_leading_zeros for details.

Usage

infer_types_count_leading_zeros(operand)

hlo_count_leading_zeros(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Infer types for custom call

Description

Infer the output types for a custom call operation.

Usage

infer_types_custom_call(
  ...,
  call_target_name,
  api_version,
  has_side_effect,
  backend_config,
  output_types,
  operand_layouts,
  result_layouts
)

Arguments

...

Input values.
call_target_name

(character(1))
The name of the custom function to call.
api_version

(integer(1))
The API version.
has_side_effect

(logical(1))
Whether the custom call has side effects.
backend_config

(list | NULL)
Optional backend configuration as a named list.
output_types

(list of ValueType | NULL)
The output types of the custom call. Default is NULL (no outputs).
operand_layouts, result_layouts

Layouts (not used for type inference).

Value

(ValueTypes)
The output types (empty for side-effect only calls).

Digamma Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlodigamma_chlodigammaop for details.

Usage

infer_types_digamma(operand)

hlo_digamma(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Divide Operator

Description

See https://openxla.org/stablehlo/spec#divide for details.

Usage

infer_types_divide(lhs, rhs)

hlo_divide(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

DotGeneral Operator

Description

See https://openxla.org/stablehlo/spec#dot_general for details.

Usage

infer_types_dot_general(lhs, rhs, dot_dimension_numbers)

hlo_dot_general(lhs, rhs, contracting_dims, batching_dims = NULL)

Arguments

lhs, rhs, contracting_dims, batching_dims

(FuncValue)
dot_dimension_numbers

(DotDimensionNumbers)
The dot dimension number.

Value

FuncValue

DynamicSlice Operator

Description

See https://openxla.org/stablehlo/spec#dynamic_slice for details.

Usage

infer_types_dynamic_slice(operand, ..., slice_sizes)

hlo_dynamic_slice(operand, ..., slice_sizes)

Arguments

operand, ..., slice_sizes

(FuncValue)

Value

FuncValue

DynamicUpdateSlice Operator

Description

See https://openxla.org/stablehlo/spec#dynamic_update_slice for details.

Usage

infer_types_dynamic_update_slice(operand, update, ...)

hlo_dynamic_update_slice(operand, update, ...)

Arguments

operand, update, ...

(FuncValue)

Value

FuncValue

Erf Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloerf_chloerfop for details.

Usage

infer_types_erf(operand)

hlo_erf(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

ErfInv Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloerf_inv_chloerf_invop for details.

Usage

infer_types_erf_inv(operand)

hlo_erf_inv(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Erfc Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chloerfc_chloerfcop for details.

Usage

infer_types_erfc(operand)

hlo_erfc(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Exponential Operator

Description

See https://openxla.org/stablehlo/spec#exponential for details.

Usage

infer_types_exponential(operand)

hlo_exponential(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

ExponentialMinusOne Operator

Description

See https://openxla.org/stablehlo/spec#exponential_minus_one for details.

Usage

infer_types_exponential_minus_one(operand)

hlo_exponential_minus_one(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Infer types for float binary operations

Description

Infer the types for float binary operations.

Usage

infer_types_float_biv(lhs, rhs)

Arguments

lhs

(ValueType)
The left-hand side operand.
rhs

(ValueType)
The right-hand side operand.

Value

(ValueType)
The inferred type.

Infer types for float unary operations

Description

Infer the types for float unary operations.

Usage

infer_types_float_uni(operand)

Arguments

operand

(ValueType)
The operand.

Value

(ValueType)
The inferred type.

Floor Operator

Description

See https://openxla.org/stablehlo/spec#floor for details.

Usage

infer_types_floor(operand)

hlo_floor(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Gather Operator

Description

See https://openxla.org/stablehlo/spec#gather for details.

Usage

infer_types_gather(
  operand,
  start_indices,
  gather_dimension_numbers,
  slice_sizes,
  indices_are_sorted
)

hlo_gather(
  operand,
  start_indices,
  gather_dimension_numbers,
  slice_sizes,
  indices_are_sorted = FALSE
)

Arguments

operand, start_indices

(FuncValue)
gather_dimension_numbers

(GatherDimensionNumbers)
The gather dimension numbers.
slice_sizes

(integer())
The sizes of the slices to gather.
indices_are_sorted

(logical(1))
Whether indices are sorted.

Value

FuncValue

Infer types for binary operations

Description

Infer the types for binary operations.

Usage

infer_types_generic_biv(lhs, rhs)

Arguments

lhs

(ValueType)
The left-hand side operand.
rhs

(ValueType)
The right-hand side operand.

Value

(ValueType)
The inferred type.

Infer types for unary operations

Description

Infer the types for unary operations.

Usage

infer_types_generic_uni(operand)

Arguments

operand

(ValueType)
The operand.

Value

(ValueType)
The inferred type.

If Operator

Description

See https://openxla.org/stablehlo/spec#if for details.

Usage

infer_types_if(pred, true_branch, false_branch)

hlo_if(pred, true_branch, false_branch, simplify = TRUE)

Arguments

pred, true_branch, false_branch

(FuncValue)
simplify

(logical(1))
Whether to simplify results by unpacking lists of length 1 into their single element.

Value

FuncValue

Infer types for integer unary operations

Description

Infer the types for integer unary operations.

Usage

infer_types_integer_uni(operand)

Arguments

operand

(ValueType)
The operand.

Value

(ValueType)
The inferred type.

Infer types for boolean integerish operations

Description

Infer the types for integerish (bool or int) binary operations.

Usage

infer_types_integerish_biv(lhs, rhs)

Arguments

lhs

(ValueType)
The left-hand side operand.
rhs

(ValueType)
The right-hand side operand.

Value

(ValueType)
The inferred type.

Infer types for integerish unary operations

Description

Infer the types for integerish (bool or int) unary operations.

Usage

infer_types_integerish_uni(operand)

Arguments

operand

(ValueType)
The operand.

Value

(ValueType)
The inferred type.

Iota Operator

Description

See https://openxla.org/stablehlo/spec#iota for details.

Usage

infer_types_iota(iota_dimension, dtype, shape)

hlo_iota(iota_dimension, dtype, shape, func = NULL)

Arguments

iota_dimension

(integer(1))
The dimension along which to generate increasing values. Must be in range ⁠[0, rank(output))⁠.
dtype

(character(1))
The data type of the output tensor. One of: bool, i8, i16, i32, i64, ui8, ui16, ui32, ui64, f32, f64 (excluding boolean).
shape

(integer())
The shape of the output tensor.
func

(Func)
The function to add the operation to. Per default, uses the last function created with hlo_func or local_func.

Value

FuncValue

IsFinite Operator

Description

See https://openxla.org/stablehlo/spec#is_finite for details.

Usage

infer_types_is_finite(operand)

hlo_is_finite(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

IsInf Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlois_inf_chlois_infop for details.

Usage

infer_types_is_inf(operand)

hlo_is_inf(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

IsNegInf Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlois_neg_inf_chlois_neg_infop for details.

Usage

infer_types_is_neg_inf(operand)

hlo_is_neg_inf(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

IsPosInf Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlois_pos_inf_chlois_pos_infop for details.

Usage

infer_types_is_pos_inf(operand)

hlo_is_pos_inf(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Lgamma Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlolgamma_chlolgammaop for details.

Usage

infer_types_lgamma(operand)

hlo_lgamma(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Log Operator

Description

See https://openxla.org/stablehlo/spec#log for details.

Usage

infer_types_log(operand)

hlo_log(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

LogPlusOne Operator

Description

See https://openxla.org/stablehlo/spec#log_plus_one for details.

Usage

infer_types_log_plus_one(operand)

hlo_log_plus_one(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Logistic Operator

Description

See https://openxla.org/stablehlo/spec#logistic for details.

Usage

infer_types_logistic(operand)

hlo_logistic(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Maximum Operator

Description

See https://openxla.org/stablehlo/spec#maximum for details.

Usage

infer_types_maximum(lhs, rhs)

hlo_maximum(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Minimum Operator

Description

See https://openxla.org/stablehlo/spec#minimum for details.

Usage

infer_types_minimum(lhs, rhs)

hlo_minimum(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Multiply Operator

Description

See https://openxla.org/stablehlo/spec#multiply for details.

Usage

infer_types_multiply(lhs, rhs)

hlo_multiply(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Negate Operator

Description

See https://openxla.org/stablehlo/spec#negate for details.

Usage

infer_types_negate(operand)

hlo_negate(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Not Operator

Description

See https://openxla.org/stablehlo/spec#not for details.

Usage

infer_types_not(operand)

hlo_not(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Infer types for numeric binary operations

Description

Infer the types for numeric binary operations.

Usage

infer_types_numeric_biv(lhs, rhs)

Arguments

lhs

(ValueType)
The left-hand side operand.
rhs

(ValueType)
The right-hand side operand.

Value

(ValueType)
The inferred type.

Infer types for numeric unary operations

Description

Infer the types for numeric unary operations.

Usage

infer_types_numeric_uni(operand)

Arguments

operand

(ValueType)
The operand.

Value

(ValueType)
The inferred type.

Or Operator

Description

See https://openxla.org/stablehlo/spec#or for details.

Usage

infer_types_or(lhs, rhs)

hlo_or(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Pad Operator

Description

See https://openxla.org/stablehlo/spec#pad for details.

Usage

infer_types_pad(
  operand,
  padding_value,
  edge_padding_low,
  edge_padding_high,
  interior_padding
)

hlo_pad(
  operand,
  padding_value,
  edge_padding_low,
  edge_padding_high,
  interior_padding
)

Arguments

operand, padding_value, edge_padding_low, edge_padding_high, interior_padding

(FuncValue)

Value

FuncValue

Polygamma Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlopolygamma_chlopolygammaop for details.

Usage

infer_types_polygamma(n, x)

hlo_polygamma(n, x)

Arguments

n, x

(FuncValue)

Value

FuncValue

Popcnt Operator

Description

See https://openxla.org/stablehlo/spec#popcnt for details.

Usage

infer_types_popcnt(operand)

hlo_popcnt(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Power Operator

Description

See https://openxla.org/stablehlo/spec#power for details.

Usage

infer_types_power(lhs, rhs)

hlo_power(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Reduce Operator

Description

See https://openxla.org/stablehlo/spec#reduce for details.

Usage

infer_types_reduce(inputs, init_values, body, dimensions)

hlo_reduce(inputs, init_values, dimensions, body)

Arguments

inputs, init_values, dimensions, body

(FuncValue)

Value

FuncValue

ReduceWindow Operator

Description

See https://openxla.org/stablehlo/spec#reduce_window for details.

Usage

infer_types_reduce_window(
  ...,
  body,
  window_dimensions,
  window_strides,
  base_dilations,
  window_dilations,
  padding
)

hlo_reduce_window(
  inputs,
  init_values,
  window_dimensions,
  window_strides,
  base_dilations,
  window_dilations,
  padding,
  body
)

Arguments

...

(Inputs, Init values)
body

(Func)
The reduction function to apply to each window.
window_dimensions

(integer())
The size of the window in each dimension.
window_strides

(integer())
The stride of the window in each dimension.
base_dilations

(integer())
The dilation factor for the input tensor.
window_dilations

(integer())
The dilation factor for the window.
padding

(matrix)
A matrix with shape ⁠[rank, 2]⁠ specifying the padding before and after each dimension.
inputs

(list() of FuncValue)
The input tensor(s) to apply the reduction to.
init_values

(list() of FuncValue)
The initial value(s) for the reduction. Must be 0-D tensors.

Remainder Operator

Description

See https://openxla.org/stablehlo/spec#remainder for details.

Usage

infer_types_remainder(lhs, rhs)

hlo_remainder(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Reshape Operator

Description

See https://openxla.org/stablehlo/spec#reshape for details.

Usage

infer_types_reshape(operand, shape)

hlo_reshape(operand, shape)

Arguments

operand, shape

(FuncValue)

Value

FuncValue

Return Values

Description

Specifies the return values of a Func and finalize it.

Usage

infer_types_return(...)

hlo_return(..., func = .current_func())

Arguments

...

(FuncValue)
Return values. There must be at least one.
func

(Func)
The function. Default is to use .current_func().

Value

(Func)

Reverse Operator

Description

See https://openxla.org/stablehlo/spec#reverse for details.

Usage

infer_types_reverse(operand, dimensions)

hlo_reverse(operand, dimensions)

Arguments

operand, dimensions

(FuncValue)

Value

FuncValue

RngBitGenerator Operator

Description

See https://openxla.org/stablehlo/spec#rng_bit_generator for details.

Usage

infer_types_rng_bit_generator(initial_state, rng_algorithm, dtype, shape)

hlo_rng_bit_generator(
  initial_state,
  rng_algorithm = c("DEFAULT", "THREE_FRY", "PHILOX"),
  dtype,
  shape
)

Arguments

initial_state, rng_algorithm, dtype, shape

(FuncValue)

Value

FuncValue

RoundNearestAfz Operator

Description

See https://openxla.org/stablehlo/spec#round_nearest_afz for details.

Usage

infer_types_round_nearest_afz(operand)

hlo_round_nearest_afz(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

RoundNearestEven Operator

Description

See https://openxla.org/stablehlo/spec#round_nearest_even for details.

Usage

infer_types_round_nearest_even(operand)

hlo_round_nearest_even(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Rsqrt Operator

Description

See https://openxla.org/stablehlo/spec#rsqrt for details.

Usage

infer_types_rsqrt(operand)

hlo_rsqrt(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Scatter Operator

Description

See https://openxla.org/stablehlo/spec#scatter for details.

Usage

infer_types_scatter(
  inputs,
  scatter_indices,
  updates,
  scatter_dimension_numbers,
  indices_are_sorted,
  unique_indices,
  update_computation
)

hlo_scatter(
  inputs,
  scatter_indices,
  updates,
  scatter_dimension_numbers,
  indices_are_sorted = FALSE,
  unique_indices = FALSE,
  update_computation
)

Arguments

inputs, scatter_indices, updates

(FuncValue)
scatter_dimension_numbers

(ScatterDimensionNumbers)
The scatter dimension numbers.
indices_are_sorted

(logical(1))
Whether indices are sorted.
unique_indices

(logical(1))
Whether indices are unique.
update_computation

(Func)
The update computation function.

Value

FuncValue

Select Operator

Description

See https://openxla.org/stablehlo/spec#select for details.

Usage

infer_types_select(pred, on_true, on_false)

hlo_select(pred, on_true, on_false)

Arguments

pred, on_true, on_false

(FuncValue)

Value

FuncValue

ShiftLeft Operator

Description

See https://openxla.org/stablehlo/spec#shift_left for details.

Usage

infer_types_shift_left(lhs, rhs)

hlo_shift_left(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

ShiftRightArithmetic Operator

Description

See https://openxla.org/stablehlo/spec#shift_right_arithmetic for details.

Usage

infer_types_shift_right_arithmetic(lhs, rhs)

hlo_shift_right_arithmetic(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

ShiftRightLogical Operator

Description

See https://openxla.org/stablehlo/spec#shift_right_logical for details.

Usage

infer_types_shift_right_logical(lhs, rhs)

hlo_shift_right_logical(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Sign Operator

Description

See https://openxla.org/stablehlo/spec#sign for details.

Usage

infer_types_sign(operand)

hlo_sign(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Sine Operator

Description

See https://openxla.org/stablehlo/spec#sine for details.

Usage

infer_types_sine(operand)

hlo_sine(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Sinh Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlosinh_chlosinhop for details.

Usage

infer_types_sinh(operand)

hlo_sinh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Slice Operator

Description

See https://openxla.org/stablehlo/spec#slice for details.

Usage

infer_types_slice(operand, start_indices, limit_indices, strides)

hlo_slice(operand, start_indices, limit_indices, strides)

Arguments

operand, start_indices, limit_indices, strides

(FuncValue)

Value

FuncValue

Sort Operator

Description

See https://openxla.org/stablehlo/spec#sort for details.

Usage

infer_types_sort(..., dimension, is_stable, comparator)

hlo_sort(..., dimension, is_stable, comparator)

Arguments

..., dimension, is_stable, comparator

(FuncValue)

Value

FuncValue

Sqrt Operator

Description

See https://openxla.org/stablehlo/spec#sqrt for details.

Usage

infer_types_sqrt(operand)

hlo_sqrt(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Square Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlosquare_chlosquareop for details.

Usage

infer_types_square(operand)

hlo_square(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Subtract Operator

Description

See https://openxla.org/stablehlo/spec#subtract for details.

Usage

infer_types_subtract(lhs, rhs)

hlo_subtract(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Tan Operator

Description

See https://openxla.org/stablehlo/spec#tan for details.

Usage

infer_types_tan(operand)

hlo_tan(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

Tanh Operator

Description

See https://openxla.org/stablehlo/spec#tanh for details.

Usage

infer_types_tanh(operand)

hlo_tanh(operand)

Arguments

operand

(FuncValue)

Value

FuncValue

TopK Operator (CHLO)

Description

This op is from the CHLO dialect, a higher-level companion to stableHLO that is lowered to stableHLO during compilation. See https://openxla.org/stablehlo/generated/chlo#chlotop_k_chlotop_kop for details.

Usage

infer_types_top_k(operand, k)

hlo_top_k(operand, k)

Arguments

operand

(FuncValue)
Tensor of integer, unsigned integer, or floating-point type with rank >= 1.
k

(integer(1))
Number of top elements to return along the last dimension. Must satisfy ⁠1 <= k <= dim(operand, -1)⁠.

Value

FuncValue

A list() of two FuncValues: the top-k values (same dtype as operand) and their indices into the last dimension (dtype i32). Ties are broken by lower index first.

Transpose Operator

Description

See https://openxla.org/stablehlo/spec#transpose for details.

Usage

infer_types_transpose(operand, permutation)

hlo_transpose(operand, permutation)

Arguments

operand, permutation

(FuncValue)

Value

FuncValue

TriangularSolve Operator

Description

See https://openxla.org/stablehlo/spec#triangular_solve for details.

Usage

infer_types_triangular_solve(
  a,
  b,
  left_side,
  lower,
  unit_diagonal,
  transpose_a
)

hlo_triangular_solve(a, b, left_side, lower, unit_diagonal, transpose_a)

Arguments

a, b

(FuncValue)
left_side

(logical(1))
If TRUE, solve op(a) * x = b. If FALSE, solve x * op(a) = b.
lower

(logical(1))
If TRUE, use lower triangle of a. If FALSE, use upper triangle.
unit_diagonal

(logical(1))
If TRUE, assume diagonal elements of a are 1.
transpose_a

(character(1))
One of "NO_TRANSPOSE", "TRANSPOSE", or "ADJOINT".

Value

FuncValue

While Operator

Description

See https://openxla.org/stablehlo/spec#while for details.

Usage

infer_types_while(..., cond, body)

hlo_while(..., cond, body, simplify = TRUE)

Arguments

..., cond, body

(FuncValue)
simplify

(logical(1))
Whether to simplify results by unpacking lists of length 1 into their single element.

Value

FuncValue

Xor Operator

Description

See https://openxla.org/stablehlo/spec#xor for details.

Usage

infer_types_xor(lhs, rhs)

hlo_xor(lhs, rhs)

Arguments

lhs, rhs

(FuncValue)

Value

FuncValue

Module

Description

A StableHLO module containing one or more Func objects. Modules allow defining multiple named functions, where functions can call each other using hlo_call.

Note: Module uses reference semantics like Func.

Usage

Module(funcs = list())

Arguments

funcs

(list() of Func)
The functions in the module.

Value

A Module object.

Op

Description

This represents a StableHLO operation.

Usage

Op(name, inputs, outputs, signature)

Arguments

name

(OpName)
The name of the operation.
inputs

(OpInputs)
The inputs to the operation.
outputs

(OpOutputs)
The outputs of the operation.
signature

(OpSignature)
The signature of the operation.

Value

(Op)

OpInputAttr

Description

Base class for operation input attributes.

Usage

OpInputAttr(name, value, dtype)

Arguments

name

(character(1))
The name of the attribute.
value

(any)
The value of the attribute.
dtype

(tengen::DataType)
The dtype of the attribute.

Value

(OpInputAttr)

OpInputAttrs

Description

List of OpInputAttrs.

Usage

OpInputAttrs(items = list())

Arguments

items

(list() of OpInputAttr)
The attributes that can be used as inputs to operations.

Value

(OpInputAttrs)

OpInputFunc

Description

This represents a function that can be used as input to an operation.

Usage

OpInputFunc(inputs, body)

Arguments

inputs

(FuncInputs)
The inputs of the function.
body

(FuncBody)
The body of the function.

Value

(OpInputFunc)

OpInputFuncs

Description

List of OpInputFuncs.

Usage

OpInputFuncs(items = list())

Arguments

items

(list() of OpInputFunc)
The functions that can be used as inputs to operations.

Value

(OpInputFuncs)

OpInputs

Description

This represents all the inputs to an operation, including values, functions, and attributes.

Usage

OpInputs(
  values,
  funcs = OpInputFuncs(),
  attrs = OpInputAttrs(),
  custom_attrs = list()
)

Arguments

values

(OpInputValues)
The values used as inputs.
funcs

(OpInputFuncs)
The functions used as inputs.
attrs

(OpInputAttrs)
The attributes used as inputs.
custom_attrs

(list)
Custom attributes. Use this attributes that require custom formatting.

Value

(OpInputs)

OpInputValue

Description

This represents a value that can be used as input to an operation.

Usage

OpInputValue(id)

Arguments

id

(ValueId)
The id of the value.

Value

(OpInputValue)

OpInputValues

Description

List of OpInputValues.

Usage

OpInputValues(items = list())

Arguments

items

(list() of OpInputValue)
The values that can be used as inputs to operations.

Value

(OpInputValues)

OpName

Description

This represents the name of an operation, containing a mnemonic and a dialect.

Usage

OpName(mnemonic, dialect = "stablehlo")

Arguments

mnemonic

(character(1))
The mnemonic of the operation.
dialect

(character(1))
The MLIR dialect this op belongs to. Defaults to "stablehlo". Use "chlo" for CHLO ops (which lower to StableHLO during compilation).

Value

(OpName)

OpOutput

Description

This represents an output of an operation.

Usage

OpOutput(id = ValueId())

Arguments

id

(ValueId)
The id of the output.

Value

(OpOutput)

OpOutputs

Description

List of OpOutputs.

Usage

OpOutputs(items = list())

Arguments

items

(list() of OpOutput)
The outputs of an operation.

Value

(OpOutputs)

OpSignature

Description

This represents the signature of an operation, defining its input and output types.

Usage

OpSignature(input_types, output_types)

Arguments

input_types

(ValueTypes)
The types of the inputs.
output_types

(ValueTypes)
The types of the outputs.

Value

(OpSignature)

Convert R value to Constant

Description

Convert R value to Constant.

Usage

r_to_constant(value, dtype = NULL, shape, ...)

Arguments

value

(any)
The value to convert.
dtype

(character(1))
The dtype of the constant.
shape

(integer())
The shape of the constant.
...

(any) Additional arguments.

Generate string representation for object

Description

This function generates a string representation of an object. In this package, this is primarily used to convert a Func to its stableHLO string representation.

Usage

repr(x, ...)

Arguments

x

The object to generate a string representation of.
...

Additional arguments passed to the method.

Value

character(1)

ScalarAttr

Description

An attribute holding a scalar value with an associated dtype.

Usage

ScalarAttr(name, value, dtype)

Arguments

name

(character(1))
The name of the attribute.
value

(numeric(1) or logical(1))
The scalar value.
dtype

(tengen::DataType)
The dtype of the scalar (e.g., IntegerType(32), FloatType(32), BooleanType()).

Value

ScalarAttr

ScatterDimensionNumbers

Description

Represents the scatter dimension numbers.

Usage

ScatterDimensionNumbers(
  update_window_dims,
  inserted_window_dims,
  input_batching_dims = integer(),
  scatter_indices_batching_dims = integer(),
  scatter_dims_to_operand_dims,
  index_vector_dim
)

Arguments

update_window_dims

(integer())
The update window dimensions.
inserted_window_dims

(integer())
The inserted window dimensions.
input_batching_dims

(integer())
The input batching dimensions.
scatter_indices_batching_dims

(integer())
The scatter indices batching dimensions.
scatter_dims_to_operand_dims

(integer())
Maps scatter dimensions to operand dimensions.
index_vector_dim

(integer(1))
The index vector dimension.

Shape

Description

Represents the shape of a tensor.

Usage

Shape(dims = integer())

Arguments

dims

(integer())

Value

Shape

StringAttr

Description

An attribute holding a string value.

Usage

StringAttr(name, value)

Arguments

name

(character(1))
The name of the attribute.
value

(character(1))
The string value.

Value

StringAttr

TensorType

Description

Represents a tensor type with a specific data type and shape.

Usage

TensorType(dtype, shape)

Arguments

dtype

(tengen::DataType)
shape

(Shape)

Value

TensorType

Convert 0-based indices to 1-based

Description

Converts all index_vec fields in a condition object from 0-based to 1-based.

Usage

to_one_based(x, ...)

Arguments

x

Condition object with index_vec fields.
...

Additional arguments (not used).

Value

Condition object with index_vec fields incremented by 1.

ValueId

Description

This represents the name of a ValueType.

Usage

ValueId(id = NULL)

Arguments

id

(character(1) or environment)
Either a fixed name or an environment. If using an environment (default), the name will be generated automatically when calling repr(), i.e. the first value id will be ⁠%0⁠, the second ⁠%1⁠, etc..

Value

(ValueId)

ValueType

Description

This represents the type of a value.

Usage

ValueType(type, shape = NULL)

Arguments

type

The type of the value (TensorType or TokenType).
shape

The shape of the value (only used when type is character).

ValueTypes

Description

List of ValueTypes.

Usage

ValueTypes(items = list())

Arguments

items

(list() of ValueType)
The types of the values.

Value

ValueTypes