[LLVMdev] TableGen related question for the Hexagon backend

[Jakob Stoklund Olesen]

On Aug 3, 2012, at 10:20 AM, Sebastian Pop <spop at codeaurora.org> wrote:

>> If you are just working around deficiencies in LLVM's tools instead
>> of trying to improve them, there is really no point in sharing that work.
> 
> Just curious: what deficiencies in LLVM do you think we working around here?

You tell me. The td files are supposed to be the primary source for a target description. You chose to use a different source language, and auto-generate td files from that. Why?

If you have an authoritative ISA description in machine-readable form, you can use it to verify that (parts of) your td files are correct. That would be very useful. But the td files should be the primary source.

I am not claiming the TableGen language is awesome, but it is never going to get better if lib/Target evolves into 17 different ways of auto-generating td files.

Most parts of LLVM are constantly improving, including the target descriptions. The goal is to have the best possible way of describing a target for assemblers, disassemblers and code generators. Auto-generated code prevents the language from evolving.

/jakob