Make CPU inductor work with dynamic shapes

These errors were found by looking at wav2vec2

See #91719

Signed-off-by: Edward Z. Yang <ezyangmeta.com>

ghstack-source-id: 51b6d644b4e30b9137354a4fb5eece5534d3412e
Pull Request resolved: #93077

torch/_inductor/mkldnn.py

@@ -8,6 +8,8 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
+from torch import sym_int
+
 from torch._dynamo.utils import fake_mode_from_tensors
 from torch.fx.experimental.optimization import (
     matches_module_pattern,
@@ -351,7 +353,9 @@ def __init__(self, linear: nn.Module, input_size: list):
 
     def _update_module_params(self, linear, input_size):
         self.__dict__ = copy.deepcopy(linear.__dict__)
-        self.batch_size = int(reduce(lambda x, y: x * y, input_size) / input_size[-1])
+        self.batch_size = sym_int(
+            reduce(lambda x, y: x * y, input_size) / input_size[-1]
+        )
         self.packed_weight = torch.nn.Parameter(
             torch.ops.mkl._mkl_reorder_linear_weight(
                 self.weight.to_mkldnn(), self.batch_size

torch/_inductor/overrides.py

@@ -4,6 +4,7 @@
 import weakref
 
 import torch
+import torch._dynamo.config as dynamo_config
 import torch.nn as nn
 from torch import _prims
 from torch._dynamo.utils import fake_mode_from_tensors
@@ -87,7 +88,8 @@ def fuse_fx(gm: torch.fx.GraphModule, example_inputs):
     gm = remove_identity(gm)
     gm = fuse_conv_bn(gm)
     # do mkldnn fusion(conv(linear)+unary(binary)
-    gm = mkldnn_fuse_fx(gm, example_inputs)
+    if not dynamo_config.dynamic_shapes:
+        gm = mkldnn_fuse_fx(gm, example_inputs)
     return gm