Remove HashOutput (#68)

Feature can be resurrected when bandwidth exists to make it actually work

src/tensora/codegen/_type_to_c.py

@@ -2,7 +2,7 @@
 
 from functools import singledispatch
 
-from ..ir.types import Array, FixedArray, Float, HashTable, Integer, Mode, Pointer, Tensor, Type
+from ..ir.types import Array, FixedArray, Float, Integer, Mode, Pointer, Tensor, Type
 
 
 def space_variable(variable: str | None = None) -> str:
@@ -37,11 +37,6 @@ def type_to_c_mode(self: Mode, variable: str | None = None) -> str:
     return "taco_mode_t" + space_variable(variable)
 
 
-@type_to_c.register(HashTable)
-def type_to_c_hash_table(self: HashTable, variable: str | None = None) -> str:
-    return "hash_table_t" + space_variable(variable)
-
-
 @type_to_c.register(Pointer)
 def type_to_c_pointer(self: Pointer, variable: str | None = None) -> str:
     return f"{type_to_c(self.target)}* restrict" + space_variable(variable)

src/tensora/ir/types.py

@@ -8,8 +8,6 @@
     "tensor",
     "Mode",
     "mode",
-    "HashTable",
-    "hash_table",
     "Pointer",
     "Array",
     "FixedArray",
@@ -54,14 +52,6 @@ class Mode(Type):
 mode = Mode()
 
 
-@dataclass(frozen=True, slots=True)
-class HashTable(Type):
-    pass
-
-
-hash_table = HashTable()
-
-
 @dataclass(frozen=True, slots=True)
 class Pointer(Type):
     target: Type

src/tensora/iteration_graph/outputs/__init__.py

@@ -1,4 +1,3 @@
 from ._append import AppendOutput
 from ._base import Output
 from ._bucket import BucketOutput
-from ._hash import HashOutput

src/tensora/iteration_graph/outputs/_append.py

@@ -21,7 +21,6 @@
 from ..identifiable_expression import ast as ie_ast
 from ._base import Output
 from ._bucket import BucketOutput
-from ._hash import HashOutput
 
 default_array_size = Multiply(IntegerLiteral(1024), IntegerLiteral(1024))
 
@@ -151,9 +150,8 @@ def next_output(
                 )
                 return next_output, next_output.write_declarations(bucket), SourceBuilder()
             else:
-                next_output = HashOutput(self.output, self.next_layer)
-                return (
-                    next_output,
-                    next_output.write_declarations(),
-                    next_output.write_cleanup(kernel_type),
+                raise NotImplementedError(
+                    "Encountered a sparse output layer preceded by a contraction layer or a later "
+                    "output layer. This requires a hash table to store intermediate outputs, "
+                    "which is not currently implemented."
                 )