Add new backend

[minisparrow]

this may have some help.

How to retarget TVM to a new ASIC chip as a device code generator?

Thanks for your suggestions!! I learning about VTA as you suggested and it seems that DMA tasks is processed through: Lable dma task by pragma with dma_copy during scheduling Inject the lable to dma intrinsic function defined in runtime by call_extern during lowing. Am I right? By the way, the backend module (op decl & sch -> tvm ir -> apply ir_pass -> codegen) is really not easy to read for me, with complex and deep class structure. I would appreciate that if you could provide some guidanc…

[RFC] support cce target name in tvm

great! so I close PR on dlpack, and send another one on TVM

https://tvm.apache.org/docs/tutorials/language/tensorize.html#sphx-glr-tutorials-language-tensorize-py