如何用SSE4.2和AVX指令編譯TensorFlow？這是運(yùn)行腳本以檢查TensorFlow是否工作的消息：I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcublas.so.8.0 locally I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcudnn.so.5 locally I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcufft.so.8.0 locally I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcuda.so.1 locally I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcurand.so.8.0 locally W tensorflow/core/platform/cpu_feature_guard.cc:95] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations. W tensorflow/core/platform/cpu_feature_guard.cc:95] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations. I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:910] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero我注意到它提到了SSE4.2和AVX，1)什么是SSE4.2和AVX？2)這些SSE4.2和AVX如何改進(jìn)TensorFlow任務(wù)的CPU計(jì)算。3)如何利用這兩個(gè)庫進(jìn)行TensorFlow編譯？