BBVectorize, when using VTTI, should not form types that will be split.

This is needed so that perl's SHA can be compiled (otherwise
BBVectorize takes far too long to find its fixed point).

I'll try to come up with a reduced test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@166738 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp
index 61e8d735e411ce4915d398dde98947bacec68fbe..95b78322caa0f0848f2b7f1e0ab19aa13b58d718 100644 (file)
--- a/lib/Transforms/Vectorize/BBVectorize.cpp
+++ b/lib/Transforms/Vectorize/BBVectorize.cpp
@@ -817,6 +817,15 @@ namespace {
                                                  IAddressSpace);
           if (VCost > ICost + JCost)
             return false;
+
+          // FIXME: We don't want to fuse to a type that will be split, even
+          // if the two input types will also be split and there is no other
+          // associated cost. This check depends on the fact
+          // that the current implementation of getMemoryOpCost returns only
+          // the type-splitting cost.
+          if (VCost > 1)
+            return false;
+
           CostSavings = ICost + JCost - VCost;
         }
       } else {
@@ -831,6 +840,16 @@ namespace {
 
       if (VCost > ICost + JCost)
         return false;
+
+      // FIXME: We don't want to fuse to a type that will be split, even
+      // if the two input types will also be split and there is no other
+      // associated cost. This check depends on the fact
+      // that the current implementation of getMemoryOpCost returns only
+      // the type-splitting cost (and does nothing else).
+      unsigned VTypeCost = VTTI->getMemoryOpCost(I->getOpcode(), VT1, 0, 0);
+      if (VTypeCost > 1)
+        return false;
+
       CostSavings = ICost + JCost - VCost;
     }