C++: Remove rank aggregate in SsaInternals
#7525
Conversation
鈥y a rank computation anymore.
| cached | ||
| final predicate hasRankInBlock(IRBlock block, int rnk) { rnk = getRank(this, block) } | ||
| /** Holds if this definition or use has index `index` in block `block`. */ | ||
| final predicate hasIndexInBlock(IRBlock block, int index) { |
There was a problem hiding this comment.
OOI: Why has the cached annotation been removed?
There was a problem hiding this comment.
The annotation was there to force the work done by getRank to be cached, but the new predicate uses the IRBlock::getInstruction member predicate, which itself uses a cached nonmember getInstruction predicate. Those predicates will all have the same contents and column order, so in practice any use of hasIndexInBlock will use the cached getInstruction directly.
| cached | ||
| final predicate hasRankInBlock(IRBlock block, int rnk) { rnk = getRank(this, block) } | ||
| /** Holds if this definition or use has index `index` in block `block`. */ | ||
| final predicate hasIndexInBlock(IRBlock block, int index) { |
There was a problem hiding this comment.
Previously, hasRankInBlock could only hold for rnk \in [1, ...] because it delegates to getRank which uses a rank aggregate (and rank aggregates start at 1).
But hasIndexInBlock will now also hold for index in [0, ...] because it delegates to block.getInstruction which is zero-based I assume.
Isn't this a behavior change or is rnk/index only used transparently and so this does not matter?
There was a problem hiding this comment.
Good point. This isn't completely behavior preserving, but all current uses of this predicate shouldn't depend on this change for the exact reason that you mentioned :)
This is hopefully a behavior-preserving optimization馃 .
Thanks to @hvitved for suggesting this.
The text was updated successfully, but these errors were encountered: