Nim: -d:release vs -d:anger size/performance impact

[zevv]

@krux02 recently mentioned that "koch temp recently got much slower."

This seems related to 721534119000c2bd53cc72b531726a6104381222, which causes the nim binary to be significantly bigger (30% for me) and slower (up to 70% for me). This is to be expected given the change in semantics of -d:release, but I'm not sure if the resulting performance hit is intended. Should the Nim compiler itself maybe be built with -d:danger instead?

[krux02]

Well, @Araq did a change. The flag -d:release now has all runtime flags enabled. -d:danger is what was -d:release before. Since I did not have problems with the way -d:release worked, I had to change my workflow to -d:danger in order to be not affected by this change. I don't think changing the semantics of an existing flag under the hood is a very bad idea. I think this change should have gone through a proper RFC process, because it affects almost everybody.

[Araq]

We can produce much better code for integer overflow checking but it's unclear how much of a speedup this will bring. We hope to have a workable "Incremental compilation" feature within months, so compile-times will be improved quite a bit.

[bobeff]

Hi, I want to ask whether -d:danger is equivalent to -d:release -x:off after the change or there is also some other checks which should be disabled to get the same effect? I'm asking because I want to use build equivalent to -d:danger, but also I want to use the same build script for Nim versions before and after 0.20.0.

[SolitudeSF]

@bobeff just use -d:release -d:danger

[Araq]

Hi, I want to ask whether -d:danger is equivalent to -d:release -x:off after the change or there is also some other checks which should be disabled to get the same effect?

Yes, -d:danger is -d:release -x:off and -d:release -x:off works on all known Nim versions.

[genotrance]

What's the pending action for this issue?

[Araq]

Faster integer checking is available in a branch that will be merged too. Once that's done, we can close this issue.

[timotheecour]

can we close this now? I don't see any other action items and -d:danger has been there for a while. Note that many other flags can improve performance further (but it's problem dependent and can sometimes hurt performance), eg --passc:-flto + usual suspects clang optimization flags

Faster integer checking is available in a branch that will be merged too. Once that's done, we can close this issue.

was merged see https://github.com/nim-lang/Nim/pull/13626#issuecomment-597930095

We can produce much better code for integer overflow checking but it's unclear how much of a speedup this will bring

see https://github.com/nim-lang/RFCs/issues/199 for some benchmarks

[Araq]

Indeed, thanks.