i've been looking into some of the proposed improvements for DI performance outlined here, but have run into an issue i don't quite understand how to handle. specifically, if i have two SILInstructions that are known to have the same parent, is there a way to determine their ordering without walking the underlying list of instructions in the block? i've sort of assumed this should be possible since when i look at SILGen output, there are increasing ids for the instruction values[1], but i haven't determined if/how that information can be used for this purpose.
not sure if this is the correct terminology, but the things like %2, %3, etc âŠī¸
it seems i may have been mistaken in my assumption that there is a way to do this based on the fact that there are increasing 'identifier' values printed in the SILGen output. i was informed that this is a byproduct of the logic in SILPrinter which also does a traversal of all blocks/instructions to generate those values.
so, i will assume this isn't possible to do without some form of local cache derived from walking the instructions at least once.
to resurrect & continue this soliloquy â it turns out there is now a utility to effectively do this, which was recently added. it will still have to walk basic blocks once to compute the indices, but IIUC it stores them efficiently within the instruction itself without requiring a new field (correct me if i'm wrong). i note this primarily out of personal interest, but it may also be useful for posterity.