[metal] Add MetalBackend method overrides for codegen

helion/_compiler/backend.py

@@ -2400,11 +2400,85 @@ def library_imports(self) -> dict[str, str]:
             ),
         }
 
+    def index_type_str(self, index_dtype: torch.dtype) -> str:
+        return "uint"
+
+    def inline_constexpr(self, name: str, value: str) -> str:
+        return f"{name} = {value}"
+
     def cast_expr(self, expr_str: str, dtype_str: str) -> str:
         return f"static_cast<{dtype_str}>({expr_str})"
 
+    def program_id_expr(self, dim: int, *, index_dtype: str) -> str:
+        return f"tgid[{dim}]"
+
+    def grid_index_expr(
+        self, offset_var: str, block_size_var: str, dtype: str, *, axis: int
+    ) -> str:
+        return f"{offset_var} + tid[{axis}]"
+
     def force_tile_mask(self) -> bool:
         return True
 
+    def full_expr(
+        self, shape_dims: list[str], value_expr: str, dtype: torch.dtype
+    ) -> str:
+        metal_type = self.dtype_str(dtype)
+        return f"{metal_type}({value_expr})"
+
+    def reshape_expr(self, expr: str, shape: str) -> str:
+        return expr
+
+    def broadcast_to_expr(self, expr: str, shape: str) -> str:
+        return expr
+
+    def zeros_expr(self, shape: str, dtype: str) -> str:
+        return "0"
+
+    def where_expr(self, mask: str, true_val: str, false_val: str) -> str:
+        return f"({mask} ? {true_val} : {false_val})"
+
+    def minimum_expr(self, a: str, b: str) -> str:
+        return f"min({a}, {b})"
+
     def supports_config_key(self, key: str) -> bool:
         return key in self._SUPPORTED_CONFIG_KEYS
+
+    def supports_precompile(self) -> bool:
+        return False
+
+    def autotune(
+        self,
+        bound_kernel: BoundKernel[Any],
+        args: Sequence[object],
+        *,
+        force: bool = True,
+        **kwargs: object,
+    ) -> Config:
+        return bound_kernel.config_spec.default_config()
+
+    def transform_host_arg(
+        self,
+        arg: Argument,
+        host_str: str,
+        tensor_host_args: list[str],
+    ) -> str:
+        """Wrap scalar SymbolArguments as 1-element tensors for buffer passing."""
+        from .device_function import SymbolArgument
+
+        if isinstance(arg, SymbolArgument):
+            device_expr = (
+                f"{tensor_host_args[0]}.device" if tensor_host_args else "'mps'"
+            )
+            return (
+                f"torch.scalar_tensor(float({host_str}), "
+                f"dtype=torch.float32, "
+                f"device={device_expr})"
+            )
+        return host_str
+
+    def launcher_keyword_args(self, config: Config, *, has_barrier: bool) -> list[str]:
+        from .device_function import DeviceFunction
+
+        dims = tuple(DeviceFunction.current().codegen.max_thread_block_dims)
+        return [f"_block_dims=({dims[0]}, {dims[1]}, {dims[2]})"]