[PATCH] D35663: [Polly] Remove dependency of `Scop::getStmtFor(Inst)` on `getStmtFor(BB)`. [NFC].

[Nandini Singhal via Phabricator via llvm-commits]

nandini12396 added inline comments.


================
Comment at: lib/Analysis/ScopInfo.cpp:4828
   StmtMap[BB].push_back(Stmt);
+  for (Instruction &Inst : *BB) {
+    assert(!InstStmtMap[&Inst] &&
----------------
nandini12396 wrote:
> Meinersbur wrote:
> > To fix the `fx_basic_util_crash.ll` case, only map the instructions that are actually in the statement (The `Instructions` vector).
> > 
> > However, this causes 31 other tests to break, particularly on `assert(Parent.getStmtFor(AccessVal) == this)`. That assertion can be removed, it is suficient if the value is syntesizable in the statement.
> > 
> > Then there are still 3 other failing tests remaining.
> yes, I had earlier inserted only the instructions of the `Instructions` vector into the Map. But I reverted because I didnt understand how can this be a non-functional change? 
> Apart from that assertion for me, only 2 other tests fail on `!Inst->mayReadOrWriteMemory()`. Im thinking why so..
> And Im also thinking why this should fix the `fx_basic_util_crash.ll` case.
Oh, I see. Sorry, I missed your comment about the test case where you said "We should probably not try to verify the arguments of synthesizable instructions." 


https://reviews.llvm.org/D35663