[Enhancement] CPU Backend #1

The old issue belong to microsoft/TileIR#9

Our core component is built on Apache TVM, which provides robust support for multiple backends (e.g., CUDA, Vulkan, OpenCL, etc.). However, currently, our project lacks a CPU backend. Given that TVM’s modular design makes it straightforward to support multiple backends, introducing a CPU backend would be a logical and valuable enhancement to our system.

Where pull request microsoft/TileIR#17 support cpu c source code generation, but we still have some todo items:

• wrap the cpp source code into a runtime to be invoked with torch tensor (torch cpp extension is recommended)