Deeploy.TilingExtension.CodeTransformationPasses.SingleBufferingTilingCodeGeneration.SingleBufferingTilingCodeGeneration

class Deeploy.TilingExtension.CodeTransformationPasses.SingleBufferingTilingCodeGeneration.SingleBufferingTilingCodeGeneration(externalMemory: str, localMemory: str, dma: AsyncDma)

Bases: TilingCodeGeneration

Methods

__init__(externalMemory: str, localMemory: str, dma: AsyncDma)

`__init__`(externalMemory, localMemory, dma)
`apply`(ctxt, executionBlock, name[, verbose])	Apply the CodeTransformation to an ExecutionBlock
`dereferenceVars`(template, varNames)
`extractDynamicReferences`(ctxt[, ...])
`generateAllTilingCode`(executionBlock, ...)
`generateLoopCode`(executionBlock, metaInfo, ...)
`generateSetupAndTeardownCode`(executionBlock, ...)
`generateTilingLoop`(ctxt, executionBlock, ...)
`indexVars`(template, varNames, index)
`isFinalMemoryLevel`(tensorMemoryConstraint)

Attributes

`parseTreeDict`
`prefix`

apply(ctxt: NetworkContext, executionBlock: ExecutionBlock, name: str, verbose: CodeGenVerbosity = CodeGenVerbosity(tilingProfiling=None, untiledProfiling=None)) → Tuple[NetworkContext, ExecutionBlock]

Apply the CodeTransformation to an ExecutionBlock

Parameters:

ctxt (NetworkContext) – Current NetworkContext
executionBlock (ExecutionBlock) – ExecutionBlock whose code you’d like to transform
name (str) – Graph node name of the operator being targetted
verbose (CodeGenVerbosity) – Control the verbosity of code generation

Returns:

Updated NetworkContext and ExecutionBlock

Return type:

Tuple[NetworkContext, ExecutionBlock]