Remove an invalid assert in TargetTransformImpl

getCastInstrCost had an assert prohibiting scalar to vector casts. Such casts,
however, are allowed. This should make the vectorizer buildbot happier.

llvm-svn: 166998

diff --git a/llvm/lib/Target/TargetTransformImpl.cpp b/llvm/lib/Target/TargetTransformImpl.cpp
index 657491b4959e2abcb061d624c803911c4e8cfd6d..27877a9320abeda9b68c99e50fb7748be005ef16 100644 (file)
--- a/llvm/lib/Target/TargetTransformImpl.cpp
+++ b/llvm/lib/Target/TargetTransformImpl.cpp
@@ -196,7 +196,6 @@ unsigned VectorTargetTransformImpl::getBroadcastCost(Type *Tp) const {
 
 unsigned VectorTargetTransformImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
                                   Type *Src) const {
-  assert(Src->isVectorTy() == Dst->isVectorTy() && "Invalid input types");
   int ISD = InstructionOpcodeToISD(Opcode);
   assert(ISD && "Invalid opcode");
 
@@ -217,6 +216,8 @@ unsigned VectorTargetTransformImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
     }
   }
 
+  unsigned ScalarizationCost = 1;
+
   // Otherwise, assume that the cast is scalarized.
   if (Dst->isVectorTy()) {
     unsigned Num = Dst->getVectorNumElements();
@@ -224,11 +225,19 @@ unsigned VectorTargetTransformImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
                                      Dst->getScalarType());
     // return the cost of multiple scalar invocation plus the cost of inserting
     // and extracting the values.
-    return getScalarizationOverhead(Dst, true, true) + Num * Cost;
+    ScalarizationCost *= getScalarizationOverhead(Dst, true, true) + Num * Cost;
   }
 
-  // Unknown scalar opcode.
-  return 1;
+  if (Src->isVectorTy()) {
+    unsigned Num = Src->getVectorNumElements();
+    unsigned Cost = getCastInstrCost(Opcode, Dst->getScalarType(),
+                                     Src->getScalarType());
+    // return the cost of multiple scalar invocation plus the cost of inserting
+    // and extracting the values.
+    ScalarizationCost *= getScalarizationOverhead(Src, true, true) + Num * Cost;
+  }
+
+  return ScalarizationCost;
 }
 
 unsigned VectorTargetTransformImpl::getCFInstrCost(unsigned Opcode) const {

diff --git a/llvm/test/Transforms/BBVectorize/X86/vs-cast.ll b/llvm/test/Transforms/BBVectorize/X86/vs-cast.ll
new file mode 100644 (file)
index 0000000..be3efca
--- /dev/null
+++ b/llvm/test/Transforms/BBVectorize/X86/vs-cast.ll
@@ -0,0 +1,12 @@
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -bb-vectorize -S | FileCheck %s
+
+define void @main() nounwind uwtable {
+entry:
+  %0 = bitcast <2 x i64> undef to i128
+  %1 = bitcast <2 x i64> undef to i128
+  ret void
+; CHECK: @main
+}
+