enabling woq fusion path on acl

.ci/pytorch/test.sh

@@ -1624,6 +1624,7 @@ test_linux_aarch64() {
        inductor/test_split_cat_fx_passes inductor/test_compile inductor/test_torchinductor \
        inductor/test_torchinductor_codegen_dynamic_shapes inductor/test_torchinductor_dynamic_shapes inductor/test_memory \
        inductor/test_triton_cpu_backend inductor/test_triton_extension_backend inductor/test_mkldnn_pattern_matcher inductor/test_cpu_cpp_wrapper \
+       inductor/test_cpu_select_algorithm \
        --shard "$SHARD_NUMBER" "$NUM_TEST_SHARDS" --verbose
 }
 

test/inductor/test_mkldnn_pattern_matcher.py

@@ -4140,9 +4140,7 @@ class TestPatternMatcher(TestPatternMatcherBase):
             s = torch.randn(s_shape, dtype=torch.bfloat16)
 
             def matcher_check_fn():
-                self.assertEqual(
-                    counters["inductor"]["woq_matcher_count"], 0 if TEST_ACL else 1
-                )
+                self.assertEqual(counters["inductor"]["woq_matcher_count"], 1)
 
             self._test_common(
                 mod,

torch/_inductor/fx_passes/mkldnn_fusion.py

@@ -1524,17 +1524,20 @@ if torch._C._has_mkldnn:
     def _mkldnn_fusion_init():
         # TODO: aarch64: enable op fusion for acl once it supports fused operators. Disabling it for now.
         # Otherwise even the matmul or innerproduct can not be accelerated with acl
-        if (
-            torch.backends.mkldnn.enabled
-            and torch.backends.mkldnn.is_available()
-            and not torch.ops.mkldnn._is_mkldnn_acl_supported()
-        ):
+        mkldnn_ready = (
+            torch.backends.mkldnn.enabled and torch.backends.mkldnn.is_available()
+        )
+        if not mkldnn_ready:
+            return
+
+        if not torch.ops.mkldnn._is_mkldnn_acl_supported():
             _register_unary_fusion()
             _register_inplace_fusion()
             _register_binary_unary_fusion()
             _register_binary_fusion()
             _register_quantization_lowerings()
-            _register_woq_lowerings()
+
+        _register_woq_lowerings()
 
     @functools.cache
     def _mkldnn_weight_pack_init():