[slimtensor] Add SlimTensor-based aoti_torch_empty_strided

Pull Request resolved: #16447

Add SlimTensor-based implementations of AOTI shim functions for tensor creation:

 `aoti_torch_create_tensor_from_blob_v2()` - Creates a non-owning SlimTensor that wraps existing memory using the `from_blob()` factory

Both functions support CPU and CUDA devices and handle all 7 SlimTensor dtypes.

Changes:
- Add `memory_slim.h` and `memory_slim.cpp` with SlimTensor-based shim implementations
- Add `runtime_shims_slim` library target to TARGETS with `CUDA_AVAILABLE=1` preprocessor flag
- Add `cuda_shim_slim_cpp_unittest()` function for SlimTensor test targets
ghstack-source-id: 336216554
@exported-using-ghexport

Differential Revision: [D90126244](https://our.internmc.facebook.com/intern/diff/D90126244/)

Author: Gasoonjia

backends/cuda/runtime/shims/memory_slim.cpp

@@ -23,6 +23,7 @@ using c10::ScalarType;
 using executorch::backends::aoti::slim::empty_strided;
 using executorch::backends::aoti::slim::from_blob;
 using executorch::backends::aoti::slim::IntArrayRef;
+using executorch::backends::aoti::slim::makeArrayRef;
 
 extern "C" {
 
@@ -76,6 +77,51 @@ AOTITorchError aoti_torch_create_tensor_from_blob_v2(
   return Error::Ok;
 }
 
+AOTITorchError aoti_torch_empty_strided(
+    int64_t ndim,
+    const int64_t* sizes_ptr,
+    const int64_t* strides_ptr,
+    int32_t dtype,
+    int32_t device_type,
+    int32_t device_index,
+    Tensor** ret_new_tensor) {
+  ET_CHECK_OR_RETURN_ERROR(
+      ret_new_tensor != nullptr,
+      InvalidArgument,
+      "aoti_torch_empty_strided: ret_new_tensor is null");
+
+  ET_CHECK_OR_RETURN_ERROR(
+      !(sizes_ptr == nullptr && ndim > 0),
+      InvalidArgument,
+      "aoti_torch_empty_strided: sizes_ptr is null but ndim > 0");
+
+  IntArrayRef sizes(sizes_ptr, static_cast<size_t>(ndim));
+
+  // Handle nullptr strides by computing contiguous strides
+  if (strides_ptr == nullptr) {
+    std::vector<int64_t> contig_strides =
+        executorch::backends::aoti::slim::compute_contiguous_strides(sizes);
+    *ret_new_tensor = new Tensor(empty_strided(
+        sizes,
+        makeArrayRef(contig_strides),
+        static_cast<ScalarType>(dtype),
+        Device(
+            static_cast<DeviceType>(device_type),
+            static_cast<DeviceIndex>(device_index))));
+  } else {
+    IntArrayRef strides(strides_ptr, static_cast<size_t>(ndim));
+    *ret_new_tensor = new Tensor(empty_strided(
+        sizes,
+        strides,
+        static_cast<ScalarType>(dtype),
+        Device(
+            static_cast<DeviceType>(device_type),
+            static_cast<DeviceIndex>(device_index))));
+  }
+
+  return Error::Ok;
+}
+
 } // extern "C"
 
 } // namespace executorch::backends::cuda

backends/cuda/runtime/shims/memory_slim.h

@@ -57,6 +57,28 @@ AOTI_SHIM_EXPORT AOTITorchError aoti_torch_create_tensor_from_blob_v2(
     const uint8_t* opaque_metadata,
     int64_t opaque_metadata_size);
 
+/**
+ * Creates an uninitialized tensor with specified dimensions, strides, and
+ * dtype on either CPU or CUDA device.
+ *
+ * @param ndim Number of dimensions in the tensor
+ * @param sizes_ptr Pointer to array of dimension sizes
+ * @param strides_ptr Pointer to array of strides for each dimension
+ * @param dtype Data type identifier (matches PyTorch scalar types)
+ * @param device_type Device type (0=CPU, 1=CUDA)
+ * @param device_index Device index
+ * @param ret_new_tensor Output parameter for the created tensor
+ * @return AOTITorchError error code (Error::Ok on success)
+ */
+AOTI_SHIM_EXPORT AOTITorchError aoti_torch_empty_strided(
+    int64_t ndim,
+    const int64_t* sizes_ptr,
+    const int64_t* strides_ptr,
+    int32_t dtype,
+    int32_t device_type,
+    int32_t device_index,
+    Tensor** ret_new_tensor);
+
 } // extern "C"
 
 } // namespace executorch::backends::cuda

backends/cuda/runtime/shims/tests/targets.bzl

@@ -71,4 +71,5 @@ def define_common_targets():
     cuda_shim_cpp_unittest("aoti_torch_assign_tensors_out")
 
     # SlimTensor-based shim tests
+    cuda_shim_slim_cpp_unittest("aoti_torch_empty_strided")
     cuda_shim_slim_cpp_unittest("aoti_torch_create_tensor_from_blob_v2")