[PATCH] D37424: [ForwardOp] Remove read accesses for all instructions that have been moved

[Michael Kruse via Phabricator via llvm-commits]

Meinersbur added a comment.

Forwarded instructions may have uses remaining in the same statement, removing the MemoryAccess will leave it without a value to use and crash in codegen. This is why removing access has to be deferred to -polly-simplify.

The test-case forward_load_tripleuse doesn't crash in codegen, even with a -polly-simplify inserted. However, the result for func2 looks strange. I'll take a look.


https://reviews.llvm.org/D37424