onnxruntime-gpu 1.7 出现的警告“Force fallback to CPU execution for node: Gather_191”等

使用 onnxruntime-gpu（简称ORT）1.7 推断 onnx模型时出现如下警告，

2021-12-01 15:50:30.792327215 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Gather_191
2021-12-01 15:50:30.792374122 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Unsqueeze_363
2021-12-01 15:50:30.792386144 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Gather_188
2021-12-01 15:50:30.792413728 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Unsqueeze_362
2021-12-01 15:50:30.792442104 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Gather_182
2021-12-01 15:50:30.792453046 [W:onnxruntime:Default, fallback_cpu_capability.cc:135 GetCpuPreferredNodes] Force fallback to CPU execution for node: Unsqueeze_361

ORT 选择 cpu 去做，是由于这个操作 相对于 cuda 在 CPU 上做效率更高。

参考：Force fallback to CPU execution for Gather, Unsqueeze, Concat nodes - onnxruntime-gpu 1.7.0, opset 12 and 13 #7792

升级 ORT 到 1.8，出现了莫名其妙的错误：

[E:onnxruntime:, sequential_executor.cc:339 Execute] Non-zero status code returned while running Conv node. Name:'Conv_2' Status Message: /onnxruntime_src/onnxruntime/core/framework/bfc_arena.cc:330 void* onnxruntime::BFCArena::AllocateRawInternal(size_t, bool) Failed to allocate memory for requested buffer of size 409856

升级到 1.9，就没问题了。