bump iree to 20240306.822 and use bundled fx_importer (#518)

Author: dan-garvey

core/iree-requirements.txt

@@ -1,2 +1,2 @@
-iree-compiler==20240226.813
-iree-runtime==20240226.813
+iree-compiler==20240306.822
+iree-runtime==20240306.822

core/shark_turbine/aot/builtins/jittable.py

@@ -26,14 +26,7 @@
     tree_unflatten,
 )
 
-# TODO: Switch to upstream fx_importer vs local fork when ready.
-# from iree.compiler.extras.fx_importer import (
-#     GraphNodeImporter,
-#     FxImporter,
-#     FxImporterHooks,
-# )
-
-from ...importers.fx_importer import (
+from iree.compiler.extras.fx_importer import (
     GraphNodeImporter,
     FxImporter,
     FxImporterHooks,

core/shark_turbine/aot/support/ir_utils.py

@@ -13,7 +13,7 @@
 import numpy as np
 import torch
 
-from ...importers.fx_importer import (
+from iree.compiler.extras.fx_importer import (
     ContextCache,
     Empty,
     EmptyType,

core/shark_turbine/dynamo/backends/cpu.py

@@ -33,7 +33,7 @@
     VmModule,
 )
 
-from ...importers.fx_importer import FxImporter
+from iree.compiler.extras.fx_importer import FxImporter
 
 import torch
 from torch._dynamo.backends.common import aot_autograd

core/shark_turbine/dynamo/tensor.py

@@ -51,7 +51,7 @@
 from iree.compiler.api import Session, Output
 from iree.compiler.passmanager import PassManager
 
-from ..importers.fx_importer import FxImporter
+from iree.compiler.extras.fx_importer import FxImporter
 
 DEFAULT_COMPILER_FLAGS = ("--iree-input-type=torch",)