The current system of MIR opt levels 0-4 is suboptimal, in part because there is confusion what the individual levels mean. As we continue to add new MIR optimizations and revise existing ones, we should take the chance to improve the opt level system.
One potential option is to remove the levels entirely and derive what passes and (and to what strength or complexity) from other options the user has provided such as -Copt-level and -Cdebuginfo. If there are additional axis that are important to capture, other flags could be introduced for those.
Another option is to redefine the opt levels. Having four levels is confusing as it doesn't intuitively map to the 0-2 levels provided by -Copt-level (although that flag also has s and z opt levels which are not present in MIR opt levels at all). We could decide to redefine MIR opt levels to be 0-2 based. One challenge here is defining what the levels themselves mean (when does an optimization fall into opt level 1 or opt level 2?).
This issue corresponds to a meeting proposal for the compiler team
steering meeting. It corresponds to a possible topic of
discussion. You can read more about the steering meeting procedure
here.
Comment policy
These issues are meant to be used as an "announcements channel"
regarding the proposal, and not as a place to discuss the technical
details. Feel free to subscribe to updates. We'll post comments when
reviewing the proposal in meetings or making a scheduling decision.
In the meantime, if you have questions or ideas, ping the proposers
on Zulip (or elsewhere).
scottmcm
commented
1 year ago
Meeting proposal info
Summary
The current system of MIR opt levels 0-4 is suboptimal, in part because there is confusion what the individual levels mean. As we continue to add new MIR optimizations and revise existing ones, we should take the chance to improve the opt level system.
One potential option is to remove the levels entirely and derive what passes and (and to what strength or complexity) from other options the user has provided such as
-Copt-leveland-Cdebuginfo. If there are additional axis that are important to capture, other flags could be introduced for those.Another option is to redefine the opt levels. Having four levels is confusing as it doesn't intuitively map to the
0-2levels provided by-Copt-level(although that flag also hassandzopt levels which are not present in MIR opt levels at all). We could decide to redefine MIR opt levels to be0-2based. One challenge here is defining what the levels themselves mean (when does an optimization fall into opt level 1 or opt level 2?).Prior discussions:
About this issue
This issue corresponds to a meeting proposal for the compiler team steering meeting. It corresponds to a possible topic of discussion. You can read more about the steering meeting procedure here.
Comment policy
These issues are meant to be used as an "announcements channel" regarding the proposal, and not as a place to discuss the technical details. Feel free to subscribe to updates. We'll post comments when reviewing the proposal in meetings or making a scheduling decision. In the meantime, if you have questions or ideas, ping the proposers on Zulip (or elsewhere).