I knew what is barrier for in DX12, but I don't know what situation that resource hazard will occur.

For example, I have a command list that consists of two command.

1. Write data to resource R
2. Read data from resource R

If the GPU will execute the command list in sequence (1 before 2), why I need barrier between 1 and 2 ?

After searching on Internet, This is because GPU has so many shader cores to process a single write command.

There are a few main reasons a barrier is needed for read-after-write situation:

• Ensuring that all pending threads/work items have actually run to completion, like you mentioned
• Making sure that non-coherent caches are flushed
• The resource might need to be decompressed in order to be read by a particular stage of the pipeline
• Some operations may be deferred until the resource is being read, for instance a clear operation may only end up happening for pixels that weren't written to

I wrote a whole series on my blog that gets into a lot of the details of barriers.