[AOTI][refactor] Remove model_container_runner_cuda.cpp (pytorch#116113)

Differential Revision: [D52301272](https://our.internmc.facebook.com/intern/diff/D52301272)
Pull Request resolved: pytorch#116113
Approved by: https://github.com/khabinov
ghstack dependencies: pytorch#116047

Author: desertfire

build_variables.bzl

@@ -652,7 +652,6 @@ libtorch_cuda_core_sources = [
     "torch/csrc/CudaIPCTypes.cpp",
     "torch/csrc/cuda/comm.cpp",
     "torch/csrc/cuda/memory_snapshot.cpp",
-    "torch/csrc/inductor/aoti_runner/model_container_runner_cuda.cpp",
     "torch/csrc/inductor/aoti_torch/shim_cuda.cpp",
     "torch/csrc/jit/codegen/fuser/cuda/fused_kernel.cpp",
     "torch/csrc/profiler/stubs/cuda.cpp",

torch/csrc/inductor/aoti_runner/model_container_runner_cpu.h

@@ -9,6 +9,10 @@ class TORCH_API AOTIModelContainerRunnerCpu : public AOTIModelContainerRunner {
       const std::string& model_so_path,
       size_t num_models = 1)
       : AOTIModelContainerRunner(model_so_path, num_models, true, "") {}
+
+  std::vector<at::Tensor> run(std::vector<at::Tensor>& inputs) {
+    return AOTIModelContainerRunner::run(inputs);
+  }
 };
 
 } // namespace torch::inductor

torch/csrc/inductor/aoti_runner/model_container_runner_cuda.cpp

@@ -1,6 +1,6 @@
 #pragma once
 
-#include <cuda_runtime_api.h>
+#include <c10/cuda/CUDAStream.h>
 #include <torch/csrc/inductor/aoti_runner/model_container_runner.h>
 
 namespace torch::inductor {
@@ -15,7 +15,13 @@ class TORCH_API AOTIModelContainerRunnerCuda : public AOTIModelContainerRunner {
 
   std::vector<at::Tensor> run(
       std::vector<at::Tensor>& inputs,
-      cudaStream_t cuda_stream_handle = nullptr);
+      cudaStream_t cuda_stream_handle = nullptr) {
+    if (cuda_stream_handle == nullptr) {
+      cuda_stream_handle = c10::cuda::getCurrentCUDAStream().stream();
+    }
+    return AOTIModelContainerRunner::run(
+        inputs, reinterpret_cast<AOTInductorStreamHandle>(cuda_stream_handle));
+  }
 };
 
 } // namespace torch::inductor