search for: pr17288

...olved. Specifically, I think that these issues should be resolved prior to the release of 3.4: PR18068 [BasicAA] PR18067 [GVN] PR16431 (multiple; I suspect covered by the previous two) PR17638 [IndVarSimplify] PR18223 [IndVarSimplify] PR17473 [LSR] PR18165 [LSR] PR16729 [Loop Vectorizer] PR17288 [Loop Vectorizer] PR18102 [x86 backend or CodeGen] PR18000 [x86 backend or CodeGen] PR17504 [x86 backend or CodeGen] PR17794 PR17073 PR16108 PR18042 [LCSSA] (This hits an assert, but may miscompile with NDEBUG) Of course, the same underlying bug may be causing more than one of these, and...

The LLVM 3.4 branches are now frozen. We’re only accepting major, super horrible bug fixes from now on. The testers are going to do a third phase of testing, but it’s mostly to verify that we don’t have any major problems left. Share and enjoy! -bw

Hi Arnold, Nadav, I've been taking a look at the preamble and bailout tests created by the loop vectorizer, and I can't help but feel it could be rather conservative. I'm not a vectorization expert, so I apologise in advance if say something obviously wrong... I'm looking in particular at the overflow check and the trip count computation. From my reading, it goes something like:

...issues should be resolved prior to the release of 3.4: > > PR18068 [BasicAA] > PR18067 [GVN] > PR16431 (multiple; I suspect covered by the previous two) > PR17638 [IndVarSimplify] > PR18223 [IndVarSimplify] > PR17473 [LSR] > PR18165 [LSR] > PR16729 [Loop Vectorizer] > PR17288 [Loop Vectorizer] > PR18102 [x86 backend or CodeGen] > PR18000 [x86 backend or CodeGen] > PR17504 [x86 backend or CodeGen] > PR17794 > PR17073 > PR16108 > > PR18042 [LCSSA] (This hits an assert, but may miscompile with NDEBUG) > > Of course, the same underlying bug...