[slimtensor] Add SlimTensor-based aoti_torch_empty_strided

backends/cuda/runtime/shims/memory_slim.cpp

@@ -76,6 +76,39 @@ 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));
+  IntArrayRef strides(strides_ptr, static_cast<size_t>(ndim));
+
+  // Create the SlimTensor using empty_strided (owning)
+  *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")