[aarch64] add mkldnn fix for torch.compile()

Author: snadampal

mkldnn_fix/onednn-cpu-aarch64-remove-weight-format-checking.patch

@@ -0,0 +1,24 @@
+cpu: aarch64: remove weight format checking
+
+with weights pre-packing enabled in torch.compile(),
+the weights come already reorderd and in oneDNN format,
+so, checking for format::any is not valid.
+---
+ src/cpu/aarch64/acl_inner_product.hpp | 1 -
+ 1 file changed, 1 deletion(-)
+
+diff --git a/src/cpu/aarch64/acl_inner_product.hpp b/src/cpu/aarch64/acl_inner_product.hpp
+index a2be164f0..eadeb92b6 100644
+--- a/src/cpu/aarch64/acl_inner_product.hpp
++++ b/src/cpu/aarch64/acl_inner_product.hpp
+@@ -101,7 +101,6 @@ struct acl_inner_product_fwd_t : public primitive_t {
+                             primitive_attr_t::skip_mask_t::post_ops, f32);
+             const bool ok = is_fwd() && !has_zero_dim_memory()
+                     && utils::one_of(true, is_fp16_ok, is_fp32_ok)
+-                    && weights_md_.format_kind == format_kind::any
+                     && set_default_params() == status::success;
+ 
+             if (!ok) return status::unimplemented;
+-- 
+2.34.1
+