skainet-compile-hlo/sk.ainet.compile.hlo

Package-level declarations

Types

ConstantFoldingPass

class ConstantFoldingPass : OptimizationPass

Constant folding optimization pass

ConversionContext

class ConversionContext(typeMapper: TypeMapper, graph: ComputeGraph? = null)

Context object for maintaining state during StableHLO conversion.

ConversionResult

sealed class ConversionResult

Result of an operation conversion

DeadCodeEliminationPass

class DeadCodeEliminationPass : OptimizationPass

Dead code elimination pass

class MlirValidator

Validator for MLIR syntax and semantic correctness.

OperationFusionPass

class OperationFusionPass : OptimizationPass

Operation fusion optimization pass

OptimizationPass

interface OptimizationPass

Interface for optimization passes

data class RegistryStats(val converterCount: Int, val operationCount: Int)

Statistics about the operation registry

StableHloConverter

class StableHloConverter(registry: StableHloOperationRegistry, typeMapper: TypeMapper, validator: MlirValidator? = null)

Main converter class that orchestrates the conversion process from ComputeGraph to StableHLO MLIR.

StableHloConverterFactory

object StableHloConverterFactory

Factory for creating StableHLO converters with default configurations.

StableHloModule

data class StableHloModule(val content: String, val functionName: String = "main", val inputSpecs: List<TensorSpec> = emptyList(), val outputSpecs: List<TensorSpec> = emptyList(), val metadata: Map<String, Any> = emptyMap())

Represents a StableHLO MLIR module text output

StableHloOperationConverter

interface StableHloOperationConverter

Interface for converting specific operations to StableHLO format.

StableHloOperationRegistry

class StableHloOperationRegistry

Registry for StableHLO operation converters.

StableHloOptimizer

class StableHloOptimizer

Optimizer for StableHLO MLIR modules.

class TypeMapper

Handles type system mapping between SKaiNET and MLIR types.

Functions

fun StableHloModule.asInputStream(): InputStream

Returns an InputStream over the MLIR content encoded as UTF-8.

fun StableHloModule.asReader(): Reader

Returns a Reader over the MLIR content.

fun toStableHlo(graph: ComputeGraph, functionName: String = "main"): StableHloModule

Export a ComputeGraph into a minimal StableHLO MLIR module text.

toStableHloLegacy

fun ~~toStableHloLegacy~~(graph: ComputeGraph, functionName: String = "main"): StableHloModule

Legacy implementation for backward compatibility.