Extracted from/depends on: https://reviews.llvm.org/D129781
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/lib/Analysis/TypeBasedAliasAnalysis.cpp | ||
---|---|---|
814 | I'm not sure changes like this add value - if we end up with a ctor that can take MD->operands() as a single argument. Is that possible already/will it be possible with some of the patches you're working on? | |
llvm/unittests/Analysis/VectorUtilsTest.cpp | ||
510 | if there is/we end up with a ctor that takes a range, maybe there should also be an assign function that does the same, that would be used here? Shape.Parameters.assign(Parameters) |
llvm/lib/Analysis/TypeBasedAliasAnalysis.cpp | ||
---|---|---|
814 | Taking into account parallel change which adds SmallVector(ArrayRef<U>) the answer is yes - NextNodes (SmallVector) can be created directly from MD->operands(). So yeah we can skip using to_vector_of here. |
llvm/unittests/Analysis/VectorUtilsTest.cpp | ||
---|---|---|
510 | I can think of extending SmallVector with range version of assign, but haven't checked yet how much value it would add excluding this particular usage. If you don't mind I would like to postpone it until rest of SmallVector related patches are submitted. |
llvm/lib/Analysis/TypeBasedAliasAnalysis.cpp | ||
---|---|---|
814 | Could you remove this/changes like this from the patch then, or do you plan to wait on changes for this patch until the ArrayRef ctor is added? (you can flag this review as "changes planned" or something) | |
llvm/unittests/Analysis/VectorUtilsTest.cpp | ||
510 | Fair enough - it is a bit unfortunate that this still causes copies & has to re-encode the small size (8 in this instance) but yeah, this is still simpler/tidier than the previous code. |
llvm/lib/Analysis/TypeBasedAliasAnalysis.cpp | ||
---|---|---|
814 |
Yes, that's my plan, I'm gonna submit https://reviews.llvm.org/D130268 first, and then rebase this change. So flagging this one as "changes planned" sounds good. |
I'm not sure changes like this add value - if we end up with a ctor that can take MD->operands() as a single argument. Is that possible already/will it be possible with some of the patches you're working on?