Always raise FailedToUnpackTile when for tile_m, tile_d in hl.tile(m, d) is used (#1009)

Author: yf225

docs/api/exceptions.md

@@ -115,6 +115,10 @@ These exceptions occur when Helion language functions are used incorrectly with
 
    Raised when tuple unpacking fails for single tile.
 
+.. autoclass:: InvalidTileRange
+
+   Raised when ``hl.tile`` is given a range where the begin exceeds the end.
+
 .. autoclass:: OverpackedTile
 
    Raised when tile is wrapped in container when indexing.

helion/exc.py

@@ -223,6 +223,13 @@ class OverpackedTile(BaseError):
     )
 
 
+class InvalidTileRange(BaseError):
+    message = (
+        "hl.tile() expects the begin of the range to be less than or equal to the end. "
+        "Got begin={0!s}, end={1!s}."
+    )
+
+
 class AssignmentMultipleTargets(NotAllowedOnDevice):
     message = "Assignment with multiple targets (a=b=1) is not allowed inside the `hl.tile` or `hl.grid` loop."
 

helion/language/loops.py

@@ -329,6 +329,11 @@ def _(
         )
     block_size_list = Tile._tiles_to_sizes(block_size_list)
 
+    if unpack:
+        target = getattr(parent, "target", None)
+        if isinstance(target, (ast.Tuple, ast.List)) and len(target.elts) > 1:
+            raise exc.FailedToUnpackTile from None
+
     results = []
     for begin_part, end_part, bs in zip(
         begin_list,
@@ -339,6 +344,8 @@ def _(
         if isinstance(begin_part, Tile) or isinstance(end_part, Tile):
             raise exc.TileOfTile
         size = end_part - begin_part  # type: ignore[operator]
+        if isinstance(size, int) and size < 0:
+            raise exc.InvalidTileRange(begin_part, end_part)
         if isinstance(size, torch.Tensor):
             size = None  # data dependent size
         if bs is None:

test/test_errors.py

@@ -135,6 +135,34 @@ def fn(x: torch.Tensor) -> torch.Tensor:
         with self.assertRaises(helion.exc.OverpackedTile):
             code_and_output(fn, (torch.randn(100, 100, device=DEVICE),))
 
+    def test_tile_invalid_range_unpack(self):
+        @helion.kernel()
+        def fn(x: torch.Tensor) -> torch.Tensor:
+            m = x.size(0)
+            m = hl.specialize(m)
+            d = x.size(2)
+            for _tile_m, _tile_d in hl.tile(m, d):
+                pass
+            return x
+
+        with self.assertRaises(helion.exc.FailedToUnpackTile):
+            code_and_output(fn, (torch.randn(192, 4, 128, device=DEVICE),))
+
+    def test_tile_invalid_range_single_dim(self):
+        @helion.kernel()
+        def fn(x: torch.Tensor) -> torch.Tensor:
+            start = hl.specialize(x.size(0))
+            end = x.size(2)
+            for _tile_m in hl.tile(start, end):
+                pass
+            return x
+
+        with self.assertRaisesRegex(
+            helion.exc.InvalidTileRange,
+            r"begin=192, end=128",
+        ):
+            code_and_output(fn, (torch.randn(192, 4, 128, device=DEVICE),))
+
     def test_invalid_config_insufficient_block_sizes(self):
         """Test that InvalidConfig shows helpful message for missing block sizes."""