Galley Sparse suitable rep

src/finchlite/autoschedule/__init__.py

@@ -29,11 +29,31 @@
     with_default_scheduler,
 )
 from .executor import LogicExecutor
-from .formatter import DefaultLogicFormatter, LogicFormatter
+from .formatter import BufferizedNDArrayFormatter, DefaultLogicFormatter, LogicFormatter
 from .loop_ordering import DefaultLoopOrderer
 from .normalize import LogicNormalizer, normalize_names
+from .optimize import DefaultLogicOptimizer
+from .rep_operations import (
+    aggregate_rep,
+    data_rep,
+    dropdims_rep,
+    eltype,
+    expanddims_rep,
+    fill_value,
+    map_rep,
+    permutedims_rep,
+)
+from .representation import (
+    DenseData,
+    ElementData,
+    ExtrudeData,
+    HollowData,
+    RepeatData,
+    SparseData,
+)
 from .stages import LogicEinsumLowerer, LogicNotationLowerer
 from .standardize import LogicStandardizer
+from .suitable_rep import SmartLogicFormatter, SuitableRep, toposort
 
 __all__ = [
     "COMPILE_NUMBA",
@@ -44,10 +64,15 @@
     "OPTIMIZE_LOGIC",
     "Aggregate",
     "Alias",
+    "BufferizedNDArrayFormatter",
     "DefaultLogicFormatter",
     "DefaultLogicOptimizer",
     "DefaultLoopOrderer",
+    "DenseData",
+    "ElementData",
+    "ExtrudeData",
     "Field",
+    "HollowData",
     "Literal",
     "LogicCapture",
     "LogicCompiler",
@@ -67,10 +92,23 @@
     "Query",
     "Relabel",
     "Reorder",
+    "RepeatData",
+    "SmartLogicFormatter",
+    "SparseData",
+    "SuitableRep",
     "Table",
     "Value",
+    "aggregate_rep",
+    "data_rep",
+    "dropdims_rep",
+    "eltype",
+    "expanddims_rep",
+    "fill_value",
     "get_default_scheduler",
+    "map_rep",
     "normalize_names",
+    "permutedims_rep",
     "set_default_scheduler",
+    "toposort",
     "with_default_scheduler",
 ]

src/finchlite/autoschedule/executor.py

@@ -8,7 +8,7 @@
 from finchlite.finch_logic.nodes import TableValue
 from finchlite.symbolic import Namespace, PostWalk, Rewrite, UnvalidatedForm
 
-from .formatter import DefaultLogicFormatter
+from .formatter import BufferizedNDArrayFormatter
 
 
 def extract_tensors(
@@ -52,7 +52,7 @@ def __init__(
         cache: bool = False,
     ):
         if ctx is None:
-            ctx = DefaultLogicFormatter()
+            ctx = BufferizedNDArrayFormatter()
         if stats_factory is None:
             stats_factory = DenseStatsFactory()
         self.ctx: LogicLoader = ctx

src/finchlite/autoschedule/formatter.py

@@ -103,3 +103,7 @@ def get_output_tns_ftype(self, fill_value: Any, shape_type: tuple[FType, ...]):
             ndim=len(shape_type),
             dimension_type=TupleFType.from_tuple(shape_type),
         )
+
+
+class BufferizedNDArrayFormatter(DefaultLogicFormatter):
+    pass