Hello I am new to the space of machine learning compilers frameworks, and have a few questions on how to generate code for my custom hardware accelarator.
Can I use TVM to generate hardware specific instruction? Lets say if I have a DMA Instruction(DMA_INST), and Matrix Multiplication Instruction (MM_INST), and other hardware specific Instructions, can TVM generate code using my custom hardware instructions, if yes where are the data structure that is used by TVM to represent such hardware specific instructions?