Deeploy.MLIRAIETypes.MLIRExecutionBlock

class Deeploy.MLIRAIETypes.MLIRExecutionBlock(computeTile: Any = None, shimTile: Any = None)

Bases: object

MLIR-specific execution state for a single operator.

Replaces the C-oriented ExecutionBlock (which holds a deque of CodeSnippet objects) with fields that carry MLIR builder state through the code-transformation pipeline.

Passes populate fields progressively:

  1. The deployer sets computeTile, shimTile, operatorRepresentation, and patternMemoryConstraint.

  2. A device-phase pass (e.g. MLIRObjectFifoPass) fills fifoMap, fifoTypes, tileSize, numTiles, kernelFuncName, and kernelObjFile.

  3. The deployer sets runtimeSequenceArgs before the runtime- sequence phase.

  4. A runtime-sequence pass (e.g. MLIRRuntimeSequencePass) reads all of the above to emit DMA configuration.

Methods

__init__(computeTile: Any = None, shimTile: Any = None) → None

__init__([computeTile, shimTile])