[llvm] [MachineCopyPropagation] Detect and fix suboptimal instruction order to enable optimizations (PR #98087)

[Quentin Colombet via llvm-commits]

qcolombet wrote:

> > Is post-RA-scheduler stable? How much work it would be to add it to all the targets from at least O3? Is there any historic reason why it is not used anywhere I don't know of and couldn't find? Is it not done yet? It generates wrong or sub-optimal code? (In my case on X86 it was interesting to me that it has just renamed some regs.)
> 
> post-RA-scheduler is the legacy scheduler. The "modern" scheduler is machine-scheduler. The legacy scheduler is still used (1) for historical reasons and (2) because it is (arguably?) better suited for DSPs (more accurate resource counting at the cost of compile time). The lack of use of dependency breaker by the new scheduler might be the third reason, but I don't think anyone considered it as a serious disadvantage.

Sorry, yes, when I said postRA scheduler, I mean the new one, but running after RA :).

> 
> > ...
> 
> I once needed tracking functionality used by MCP for a different reason, so in my opinion it is worthwile to make the tracker publicly available (regardless of the direction this patch goes).

Sounds like a plan to me.



https://github.com/llvm/llvm-project/pull/98087