fix an ASAN error in MKLDNN (pytorch#104331)

Fixes ASAN stack-use-after-scope in MKLDNN.
The stack trace is
```
2023-06-27T16:37:20.9099950Z ==1424==ERROR: AddressSanitizer: stack-use-after-scope on address 0x7f0c5dc20980 at pc 0x7f0c61286a73 bp 0x7ffef8e76990 sp 0x7ffef8e76118
2023-06-27T16:37:20.9100054Z READ of size 24 at 0x7f0c5dc20980 thread T0
2023-06-27T16:37:20.9100327Z     #0 0x7f0c61286a72 in memcmp (/usr/lib/llvm-7/lib/clang/7.0.1/lib/linux/libclang_rt.asan-x86_64.so+0x5da72)
2023-06-27T16:37:20.9100701Z     #1 0x7f0c2f395d0b in c10::ArrayRef<long>::equals(c10::ArrayRef<long>) const (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xcb8bd0b)
2023-06-27T16:37:20.9101196Z     #2 0x7f0c314a1bb1 in at::native::mkldnn_matmul(at::Tensor const&, at::Tensor const&, at::Tensor const&, float, float) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xec97bb1)
2023-06-27T16:37:20.9101714Z     ROCm#3 0x7f0c301f49c5 in at::native::bmm_out_or_baddbmm_(at::Tensor const&, at::Tensor const&, at::Tensor const&, c10::Scalar const&, c10::Scalar const&, bool) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xd9ea9c5)
2023-06-27T16:37:20.9102153Z     ROCm#4 0x7f0c301f85ab in at::native::structured_bmm_out_cpu::impl(at::Tensor const&, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xd9ee5ab)
2023-06-27T16:37:20.9102601Z     ROCm#5 0x7f0c32cb3cb6 in at::(anonymous namespace)::wrapper_CPU_bmm(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x104a9cb6)
2023-06-27T16:37:20.9103662Z     ROCm#6 0x7f0c32ea1f43 in c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (at::Tensor const&, at::Tensor const&), &(at::(anonymous namespace)::wrapper_CPU_bmm(at::Tensor const&, at::Tensor const&))>, at::Tensor, c10::guts::typelist::typelist<at::Tensor const&, at::Tensor const&> >, at::Tensor (at::Tensor const&, at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x10697f43)
2023-06-27T16:37:20.9104330Z     ROCm#7 0x7f0c3187252a in at::Tensor c10::Dispatcher::redispatch<at::Tensor, at::Tensor const&, at::Tensor const&>(c10::TypedOperatorHandle<at::Tensor (at::Tensor const&, at::Tensor const&)> const&, c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) const (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xf06852a)
2023-06-27T16:37:20.9104756Z     ROCm#8 0x7f0c3257e097 in at::_ops::bmm::redispatch(c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xfd74097)
2023-06-27T16:37:20.9105237Z     ROCm#9 0x7f0c383c31c3 in torch::autograd::VariableType::(anonymous namespace)::bmm(c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x15bb91c3)
2023-06-27T16:37:20.9106496Z     ROCm#10 0x7f0c383c25b9 in c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (c10::DispatchKeySet, at::Tensor const&, at::Tensor const&), &(torch::autograd::VariableType::(anonymous namespace)::bmm(c10::DispatchKeySet, at::Tensor const&, at::Tensor const&))>, at::Tensor, c10::guts::typelist::typelist<c10::DispatchKeySet, at::Tensor const&, at::Tensor const&> >, at::Tensor (c10::DispatchKeySet, at::Tensor const&, at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x15bb85b9)
2023-06-27T16:37:20.9106874Z     ROCm#11 0x7f0c3257da60 in at::_ops::bmm::call(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xfd73a60)
2023-06-27T16:37:20.9107275Z     ROCm#12 0x7f0c301fc0e2 in at::native::_matmul_impl(at::Tensor&, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xd9f20e2)
2023-06-27T16:37:20.9107647Z     ROCm#13 0x7f0c301f9c21 in at::native::matmul(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xd9efc21)
2023-06-27T16:37:20.9108853Z     ROCm#14 0x7f0c33dca7e3 in c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer<at::Tensor (at::Tensor const&, at::Tensor const&), &(at::(anonymous namespace)::(anonymous namespace)::wrapper_CompositeImplicitAutograd__matmul(at::Tensor const&, at::Tensor const&))>, at::Tensor, c10::guts::typelist::typelist<at::Tensor const&, at::Tensor const&> >, at::Tensor (at::Tensor const&, at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x115c07e3)
2023-06-27T16:37:20.9109255Z     ROCm#15 0x7f0c32958ef0 in at::_ops::matmul::call(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x1014eef0)
2023-06-27T16:37:20.9110023Z     ROCm#16 0x7f0c2f596b62 in at::autocast::WrapFunction_<(at::autocast::CastPolicy)0, (c10::DeviceType)0, at::Tensor (at::Tensor const&, at::Tensor const&), &(at::_ops::matmul::call(at::Tensor const&, at::Tensor const&)), at::Tensor, c10::guts::typelist::typelist<at::Tensor const&, at::Tensor const&> >::call(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xcd8cb62)
2023-06-27T16:37:20.9110723Z     ROCm#17 0x7f0c2f348403 in c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<at::Tensor (*)(at::Tensor const&, at::Tensor const&), at::Tensor, c10::guts::typelist::typelist<at::Tensor const&, at::Tensor const&> >::operator()(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xcb3e403)
2023-06-27T16:37:20.9111596Z     ROCm#18 0x7f0c2f348063 in c10::impl::wrap_kernel_functor_unboxed_<c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<at::Tensor (*)(at::Tensor const&, at::Tensor const&), at::Tensor, c10::guts::typelist::typelist<at::Tensor const&, at::Tensor const&> >, at::Tensor (at::Tensor const&, at::Tensor const&)>::call(c10::OperatorKernel*, c10::DispatchKeySet, at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0xcb3e063)
2023-06-27T16:37:20.9111976Z     ROCm#19 0x7f0c32958ef0 in at::_ops::matmul::call(at::Tensor const&, at::Tensor const&) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_cpu.so+0x1014eef0)
2023-06-27T16:37:20.9112383Z     ROCm#20 0x7f0c5803dc3e in torch::autograd::THPVariable_matmul(_object*, _object*, _object*) (/opt/conda/envs/py_3.9/lib/python3.9/site-packages/torch/lib/libtorch_python.so+0x2b2cc3e)
2023-06-27T16:37:20.9112561Z warning: parsing line table prologue at 0x00000000 should have ended at 0x0000050b but it ended at 0x0000050a
2023-06-27T16:37:20.9112713Z     ROCm#21 0x5074a6 in cfunction_call (/opt/conda/envs/py_3.9/bin/python3.9+0x5074a6)
2023-06-27T16:37:20.9112857Z     ROCm#22 0x505997 in _PyObject_Call (/opt/conda/envs/py_3.9/bin/python3.9+0x505997)
2023-06-27T16:37:20.9113114Z     ROCm#23 0x505997 in PyObject_Call /croot/python-split_1684193875530/work/build-static/<invalid>:293:12
2023-06-27T16:37:20.9113258Z     ROCm#24 0x4ed302 in do_call_core (/opt/conda/envs/py_3.9/bin/python3.9+0x4ed302)
2023-06-27T16:37:20.9113633Z     ROCm#25 0x4ed302 in _PyEval_EvalFrameDefault /croot/python-split_1684193875530/work/build-static/<invalid>:3582:22
2023-06-27T16:37:20.9113780Z     ROCm#26 0x4e6729 in _PyEval_EvalFrame (/opt/conda/envs/py_3.9/bin/python3.9+0x4e6729)
2023-06-27T16:37:20.9114041Z     ROCm#27 0x4e6729 in _PyEval_EvalCode /croot/python-split_1684193875530/work/build-static/<invalid>:4329:14
2023-06-27T16:37:20.9114202Z     ROCm#28 0x4efd7d in _PyFunction_Vectorcall (/opt/conda/envs/py_3.9/bin/python3.9+0x4efd7d)
```

Pull Request resolved: pytorch#104331
Approved by: https://github.com/soulitzer

Author: cyyever

aten/src/ATen/native/mkldnn/Matmul.cpp

@@ -181,13 +181,15 @@ void mkldnn_matmul(
   Tensor mat2_ = is_mkldnn_optimized_format(mat2_unsqueezed) ? mat2_unsqueezed : mat2_unsqueezed.contiguous();
   // Make sure mat1 and mat2 have default contiguous strides if they are contiguous tensors for better performance.
   auto mat1_sizes = mat1_.sizes();
-  IntArrayRef mat1_default_contiguous_strides = c10::contiguous_strides(mat1_sizes);
-  if (mat1_.is_contiguous() && mat1_.strides() != mat1_default_contiguous_strides) {
+  auto mat1_default_contiguous_strides = c10::contiguous_strides(mat1_sizes);
+  if (mat1_.is_contiguous() &&
+      mat1_.strides() != c10::IntArrayRef(mat1_default_contiguous_strides)) {
      mat1_ = mat1_.as_strided(mat1_sizes, mat1_default_contiguous_strides);
   }
   auto mat2_sizes = mat2_.sizes();
-  IntArrayRef mat2_default_contiguous_strides = c10::contiguous_strides(mat2_sizes);
-  if (mat2_.is_contiguous() && mat2_.strides() != mat2_default_contiguous_strides) {
+  auto mat2_default_contiguous_strides = c10::contiguous_strides(mat2_sizes);
+  if (mat2_.is_contiguous() &&
+      mat2_.strides() != c10::IntArrayRef(mat2_default_contiguous_strides)) {
     mat2_ = mat2_.as_strided(mat2_sizes, mat2_default_contiguous_strides);
   }