[PATCH] D13886: Loop Vectorizer - skiping "bitcast" allows to vectorize more loops

[Michael Zolotukhin via llvm-commits]

mzolotukhin added a comment.

Hi Elena,

The change looks reasonable, one question from me inline.

Thanks,
Michael


================
Comment at: ../test/Transforms/LoopVectorize/gep_with_bitcast.ll:28-29
@@ +27,4 @@
+  %arrayidx = getelementptr inbounds double*, double** %in, i64 %indvars.iv
+  %tmp53 = bitcast double** %arrayidx to i64*
+  %tmp54 = load i64, i64* %tmp53, align 8
+  %cmp1 = icmp eq i64 %tmp54, 0
----------------
Could it be a problem if we cast from, say, `double**` to `i8*`? Should we check the type we're pointing to when we do this?


Repository:
  rL LLVM

http://reviews.llvm.org/D13886