p-token: Tweaks to processors

[febo]

Problem

p-token processor is divided into "fast" and "normal", where the "fast" is responsible for a small subset of instructions to reduce the number of comparisons to determine which instruction to execute. Instructions that are not placed on the "fast" tend to have CU overhead, therefore popular instructions such as transfer_checked, burn and burn_checked should be placed on the "fast" path.

Solution

Move transfer_checked, burn and burn_checked to the "fast" processor. This PR also removes inline attributes of some processors where there are no gains in CUs for doing that.

[joncinque]

Looks great!

By the way, did you ever try testing with codegen-units = 1 to see if it helped CU usage or binary size at all? https://doc.rust-lang.org/rustc/codegen-options/index.html#codegen-units

[febo]

Looks great!

By the way, did you ever try testing with codegen-units = 1 to see if it helped CU usage or binary size at all? https://doc.rust-lang.org/rustc/codegen-options/index.html#codegen-units

I did – there were no differences to the CU nor binary size.

[joncinque]

Gotcha, thanks for the info!