This Patch corrects a problem whereby the optimization to use a faster divide

instruction (for Intel Atom) was not being done by Clang, because
the type context used by Clang is not the default context.

It fixes the problem by getting the global context types for each div/rem
instruction in order to compare them against the types in the BypassTypeMap.

Tests for this will be done as a separate patch to Clang.

Patch by Tyler Nowicki.

llvm-svn: 165126

1 files changed, 15 insertions, 5 deletions

diff --git a/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
index ac18b7d5a05..821b588112d 100644
--- a/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
+++ b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
@@ -238,14 +238,24 @@ bool llvm::bypassSlowDivision(Function &F,
     if (!UseDivOp && !UseRemOp)
       continue;
 
+    // Skip division on vector types, only optimize integer instructions
+    if (!J->getType()->isIntegerTy())
+      continue;
+
+    // Get same type in global context
+    IntegerType *T = cast<IntegerType>(J->getType());
+    IntegerType *GT = IntegerType::get(getGlobalContext(), T->getBitWidth());
+
     // Continue if div/rem type is not bypassed
-    DenseMap<Type *, Type *>::const_iterator BT =
-      BypassTypeMap.find(J->getType());
-    if (BT == BypassTypeMap.end())
+    DenseMap<Type *, Type *>::const_iterator BI = BypassTypeMap.find(GT);
+    if (BI == BypassTypeMap.end())
       continue;
 
-    IntegerType *BypassType = cast<IntegerType>(BT->second);
-    MadeChange |= reuseOrInsertFastDiv(F, I, J, BypassType, UseDivOp,
+    // Get the bypass type in the original context
+    IntegerType *GBT = cast<IntegerType>(BI->second);
+    IntegerType *BT = IntegerType::get(J->getContext(), GBT->getBitWidth());
+
+    MadeChange |= reuseOrInsertFastDiv(F, I, J, BT, UseDivOp,
                                        UseSignedOp, DivCache);
   }