[llvm] [MCP] Move dependencies if they block copy propagation (PR #105562)

[Gábor Spaits via llvm-commits]

spaits wrote:


> The impact is still significant on the compile time without a huge impact on the generated code.
> What is the reason for tests with faster compile time?
> 
> This is surprising to me.

I don't know. I was suprised too.

So basically the benchmarking is:
- I compile a version of llvm w my patch
- I go to the benchmark build dir run cmake
- run ninja
- run the lit command four times. Each time a different json file is produced.
- let the laptop cool down for 10 mins
- compile a version of llvm wo my patch
and the others are the sam as before

There is only one cycle that has been done with the new node restricted patch.

That result were compiled with a wo patch version that was produced before (so 4 file vs 4 file)
I saw -0.1 so I decide to dig up an older wo patchwasurement set and also compared that to the result of this measurement. There I got + 0.2.

This is why I assumed that it is not much because once its slightly positive once its slightly negative.



https://github.com/llvm/llvm-project/pull/105562