Fix for #19973
http://llvm.org/bugs/show_bug.cgi?id=19973
Fails 1 test, which seems to be a test checking for the behaviour I don't think we want.
Differential D4156
Fix for #19973 kvanberendonck on Jun 16 2014, 3:21 AM. Authored by
Details
Fix for #19973 http://llvm.org/bugs/show_bug.cgi?id=19973 Fails 1 test, which seems to be a test checking for the behaviour I don't think we want.
Diff Detail Event TimelineComment Actions Sorry for the whitespace. Originally I had the constant offset check there, but I had to move it because shouldMergeGEP wasn't doing the right thing. Comment Actions Now removed the test which was failing (testing for incorrect behaviour) and added 2 new regression test cases -- one of which currently fails (test_gep_canon_2_add). I'm working on getting it right again. Comment Actions There may be a constant floating problem in the direction we float constants outlined by the bottom tests. It's hard to pull constants out of stacks of arithmetic because they are moved to the bottom(?). It would be more efficient if they were floated to the top (i.e. closer to the things that use them), maybe. Comment Actions This patch has gone quite stale. I don't really understand why this particular change is the right one, I think it'll need some comments or other clarification in that respect before we accept it... If you end up wanting to continue working on this, try to do that, update the patch, and re-add me so I can take a look. |